Bump postgres version

2026-01-18 19:02:56 +00:00 · 2021-09-27 12:08:55 +03:00 · 2021-09-24 19:21:16 +03:00 · 2021-09-24 19:10:18 +03:00 · 2021-09-24 19:07:30 +03:00 · 2021-09-24 18:19:41 +03:00
113 changed files with 2498 additions and 5571 deletions
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -24,12 +24,6 @@ jobs:
  # A job to build postgres
  build-postgres:
    executor: zenith-build-executor
-    parameters:
-      build_type:
-        type: enum
-        enum: ["debug", "release"]
-    environment:
-      BUILD_TYPE: << parameters.build_type >>
    steps:
        # Checkout the git repo (circleci doesn't have a flag to enable submodules here)
      - checkout
@@ -45,7 +39,7 @@ jobs:
          name: Restore postgres cache
          keys:
            # Restore ONLY if the rev key matches exactly
-            - v04-postgres-cache-<< parameters.build_type >>-{{ checksum "/tmp/cache-key-postgres" }}
+            - v03-postgres-cache-{{ checksum "/tmp/cache-key-postgres" }}

        # FIXME We could cache our own docker container, instead of installing packages every time.
      - run:
@@ -65,12 +59,12 @@ jobs:
            if [ ! -e tmp_install/bin/postgres ]; then
              # "depth 1" saves some time by not cloning the whole repo
              git submodule update --init --depth 1
-              make postgres -j8
+              make postgres
            fi

      - save_cache:
          name: Save postgres cache
-          key: v04-postgres-cache-<< parameters.build_type >>-{{ checksum "/tmp/cache-key-postgres" }}
+          key: v03-postgres-cache-{{ checksum "/tmp/cache-key-postgres" }}
          paths:
            - tmp_install

@@ -102,7 +96,7 @@ jobs:
          name: Restore postgres cache
          keys:
            # Restore ONLY if the rev key matches exactly
-            - v04-postgres-cache-<< parameters.build_type >>-{{ checksum "/tmp/cache-key-postgres" }}
+            - v03-postgres-cache-{{ checksum "/tmp/cache-key-postgres" }}

      - restore_cache:
          name: Restore rust cache
@@ -241,21 +235,17 @@ jobs:
            if << parameters.run_in_parallel >>; then
              EXTRA_PARAMS="-n4 $EXTRA_PARAMS"
            fi;
-            ./netstat-script.sh &
-            NS_PID=$!
            # Run the tests.
            #
            # The junit.xml file allows CircleCI to display more fine-grained test information
            # in its "Tests" tab in the results page.
+            # -s prevents pytest from capturing output, which helps to see
+            # what's going on if the test hangs
            # --verbose prints name of each test (helpful when there are
            # multiple tests in one file)
            # -rA prints summary in the end
            # -n4 uses four processes to run tests via pytest-xdist
-            # -s is not used to prevent pytest from capturing output, because tests are running
-            # in parallel and logs are mixed between different tests
-            pipenv run pytest --junitxml=$TEST_OUTPUT/junit.xml --tb=short --verbose -rA $TEST_SELECTION $EXTRA_PARAMS
-            kill $NS_PID
-            awk '/===/ {if (count) print count; print; count=0; next} {count++} END {print count}' $TEST_OUTPUT/netstat.stdout > $TEST_OUTPUT/netstat_stats.stdout
+            pipenv run pytest --junitxml=$TEST_OUTPUT/junit.xml --tb=short -s --verbose -rA $TEST_SELECTION $EXTRA_PARAMS
      - run:
          # CircleCI artifacts are preserved one file at a time, so skipping
          # this step isn't a good idea. If you want to extract the
@@ -264,7 +254,7 @@ jobs:
          when: always
          command: |
            du -sh /tmp/test_output/*
-            find /tmp/test_output -type f ! -name "pg.log" ! -name "pageserver.log" ! -name "safekeeper.log" ! -name "regression.diffs" ! -name "junit.xml" ! -name "*.filediff" ! -name "*.stdout" ! -name "*.stderr" -delete
+            find /tmp/test_output -type f ! -name "pg.log" ! -name "pageserver.log" ! -name "wal_acceptor.log" ! -name "regression.diffs" ! -name "junit.xml" ! -name "*.filediff" -delete
            du -sh /tmp/test_output/*
      - store_artifacts:
          path: /tmp/test_output
@@ -329,7 +319,8 @@ jobs:
                \"inputs\": {
                  \"ci_job_name\": \"zenith-remote-ci\",
                  \"commit_hash\": \"$CIRCLE_SHA1\",
-                  \"remote_repo\": \"$LOCAL_REPO\"
+                  \"remote_repo\": \"$LOCAL_REPO\",
+                  \"zenith_image_branch\": \"$CIRCLE_BRANCH\"
                }
              }"

@@ -337,18 +328,14 @@ workflows:
  build_and_test:
    jobs:
      - check-codestyle
-      - build-postgres:
-          name: build-postgres-<< matrix.build_type >>
-          matrix:
-            parameters:
-              build_type: ["debug", "release"]
+      - build-postgres
      - build-zenith:
          name: build-zenith-<< matrix.build_type >>
          matrix:
            parameters:
              build_type: ["debug", "release"]
          requires:
-            - build-postgres-<< matrix.build_type >>
+            - build-postgres
      - run-pytest:
          name: pg_regress-tests-<< matrix.build_type >>
          matrix:
--- a/.dockerignore
+++ b/.dockerignore
@@ -2,17 +2,12 @@
 **/__pycache__
 **/.pytest_cache

-.git
-target
-tmp_check
-tmp_install
-tmp_check_cli
-test_output
-.vscode
-.zenith
-integration_tests/.zenith
-.mypy_cache
-
-Dockerfile
-.dockerignore
-
+/target
+/tmp_check
+/tmp_install
+/tmp_check_cli
+/test_output
+/.vscode
+/.zenith
+/integration_tests/.zenith
+/Dockerfile
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -26,21 +26,18 @@ dependencies = [
 "winapi",
 ]

-[[package]]
-name = "ansi_term"
-version = "0.12.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d52a9bb7ec0cf484c551830a7ce27bd20d67eac647e1befb56b0be4ee39a55d2"
-dependencies = [
- "winapi",
-]
-
 [[package]]
 name = "anyhow"
 version = "1.0.42"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "595d3cfa7a60d4555cb5067b99f07142a08ea778de5cf993f7b75c7d8fabc486"

+[[package]]
+name = "arc-swap"
+version = "1.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e906254e445520903e7fc9da4f709886c84ae4bc4ddaf0e093188d66df4dc820"
+
 [[package]]
 name = "async-trait"
 version = "0.1.50"
@@ -301,7 +298,7 @@ version = "2.33.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "37e58ac78573c40708d45522f0d80fa2f01cc4f9b4e2bf749807255454312002"
 dependencies = [
- "ansi_term 0.11.0",
+ "ansi_term",
 "atty",
 "bitflags",
 "strsim",
@@ -310,26 +307,6 @@ dependencies = [
 "vec_map",
 ]

-[[package]]
-name = "const_format"
-version = "0.2.21"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4556f63e28a78fa5e6f310cfea5647a25636def49a338ab69e33b34a3382057b"
-dependencies = [
- "const_format_proc_macros",
-]
-
-[[package]]
-name = "const_format_proc_macros"
-version = "0.2.21"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "552782506c398da94466b364973b563887e0ca078bf33a76d4163736165e3594"
-dependencies = [
- "proc-macro2",
- "quote",
- "unicode-xid",
-]
-
 [[package]]
 name = "control_plane"
 version = "0.1.0"
@@ -390,6 +367,26 @@ dependencies = [
 "rustc_version",
 ]

+[[package]]
+name = "crossbeam-channel"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "06ed27e177f16d65f0f0c22a213e17c696ace5dd64b14258b52f9417ccb52db4"
+dependencies = [
+ "cfg-if 1.0.0",
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "crossbeam-utils"
+version = "0.8.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d82cfc11ce7f2c3faef78d8a684447b40d503d9681acebed6cb728d45940c4db"
+dependencies = [
+ "cfg-if 1.0.0",
+ "lazy_static",
+]
+
 [[package]]
 name = "crypto-mac"
 version = "0.10.0"
@@ -428,6 +425,16 @@ dependencies = [
 "dirs-sys",
 ]

+[[package]]
+name = "dirs-next"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b98cf8ebf19c3d1b223e151f99a4f9f0690dca41414773390fc824184ac833e1"
+dependencies = [
+ "cfg-if 1.0.0",
+ "dirs-sys-next",
+]
+
 [[package]]
 name = "dirs-sys"
 version = "0.3.6"
@@ -439,6 +446,17 @@ dependencies = [
 "winapi",
 ]

+[[package]]
+name = "dirs-sys-next"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4ebda144c4fe02d1f7ea1a7d9641b6fc6b580adcfa024ae48797ecdeb6825b4d"
+dependencies = [
+ "libc",
+ "redox_users",
+ "winapi",
+]
+
 [[package]]
 name = "dlv-list"
 version = "0.2.3"
@@ -918,15 +936,6 @@ dependencies = [
 "cfg-if 1.0.0",
 ]

-[[package]]
-name = "matchers"
-version = "0.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f099785f7595cc4b4553a174ce30dd7589ef93391ff414dbb67f62392b9e0ce1"
-dependencies = [
- "regex-automata",
-]
-
 [[package]]
 name = "matches"
 version = "0.1.8"
@@ -1170,12 +1179,10 @@ dependencies = [
 "bytes",
 "chrono",
 "clap",
- "const_format",
 "crc32c",
 "daemonize",
 "futures",
 "hex",
- "hex-literal",
 "humantime",
 "hyper",
 "lazy_static",
@@ -1191,12 +1198,10 @@ dependencies = [
 "scopeguard",
 "serde",
 "serde_json",
- "signal-hook",
 "tar",
 "thiserror",
 "tokio",
 "toml",
- "tracing",
 "workspace_hack",
 "zenith_metrics",
 "zenith_utils",
@@ -1401,7 +1406,6 @@ dependencies = [
 "hex",
 "md5",
 "rand",
- "reqwest",
 "rustls",
 "serde",
 "serde_json",
@@ -1504,15 +1508,6 @@ dependencies = [
 "regex-syntax",
 ]

-[[package]]
-name = "regex-automata"
-version = "0.1.10"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6c230d73fb8d8c1b9c0b3135c5142a8acee3a0558fb8db5cf1cb65f8d7862132"
-dependencies = [
- "regex-syntax",
-]
-
 [[package]]
 name = "regex-syntax"
 version = "0.6.25"
@@ -1671,6 +1666,12 @@ dependencies = [
 "webpki",
 ]

+[[package]]
+name = "rustversion"
+version = "1.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "61b3909d758bb75c79f23d4736fac9433868679d3ad2ea7a61e3c25cfda9a088"
+
 [[package]]
 name = "ryu"
 version = "1.0.5"
@@ -1828,32 +1829,12 @@ dependencies = [
 "opaque-debug",
 ]

-[[package]]
-name = "sharded-slab"
-version = "0.1.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "740223c51853f3145fe7c90360d2d4232f2b62e3449489c207eccde818979982"
-dependencies = [
- "lazy_static",
-]
-
 [[package]]
 name = "shlex"
 version = "1.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "42a568c8f2cd051a4d283bd6eb0343ac214c1b0f1ac19f93e1175b2dee38c73d"

-[[package]]
-name = "signal-hook"
-version = "0.3.10"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9c98891d737e271a2954825ef19e46bd16bdb98e2746f2eec4f7a4ef7946efd1"
-dependencies = [
- "cc",
- "libc",
- "signal-hook-registry",
-]
-
 [[package]]
 name = "signal-hook-registry"
 version = "1.4.0"
@@ -1886,6 +1867,59 @@ version = "0.4.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f173ac3d1a7e3b28003f40de0b5ce7fe2710f9b9dc3fc38664cebee46b3b6527"

+[[package]]
+name = "slog"
+version = "2.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8347046d4ebd943127157b94d63abb990fcf729dc4e9978927fdf4ac3c998d06"
+
+[[package]]
+name = "slog-async"
+version = "2.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c60813879f820c85dbc4eabf3269befe374591289019775898d56a81a804fbdc"
+dependencies = [
+ "crossbeam-channel",
+ "slog",
+ "take_mut",
+ "thread_local",
+]
+
+[[package]]
+name = "slog-scope"
+version = "4.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2f95a4b4c3274cd2869549da82b57ccc930859bdbf5bcea0424bc5f140b3c786"
+dependencies = [
+ "arc-swap",
+ "lazy_static",
+ "slog",
+]
+
+[[package]]
+name = "slog-stdlog"
+version = "4.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8228ab7302adbf4fcb37e66f3cda78003feb521e7fd9e3847ec117a7784d0f5a"
+dependencies = [
+ "log",
+ "slog",
+ "slog-scope",
+]
+
+[[package]]
+name = "slog-term"
+version = "2.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "95c1e7e5aab61ced6006149ea772770b84a0d16ce0f7885def313e4829946d76"
+dependencies = [
+ "atty",
+ "chrono",
+ "slog",
+ "term",
+ "thread_local",
+]
+
 [[package]]
 name = "smallvec"
 version = "1.6.1"
@@ -1941,6 +1975,12 @@ dependencies = [
 "unicode-xid",
 ]

+[[package]]
+name = "take_mut"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f764005d11ee5f36500a149ace24e00e3da98b0158b3e2d53a7495660d3f4d60"
+
 [[package]]
 name = "tap"
 version = "1.0.1"
@@ -1972,6 +2012,17 @@ dependencies = [
 "winapi",
 ]

+[[package]]
+name = "term"
+version = "0.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c59df8ac95d96ff9bede18eb7300b0fda5e5d8d90960e76f8e14ae765eedbf1f"
+dependencies = [
+ "dirs-next",
+ "rustversion",
+ "winapi",
+]
+
 [[package]]
 name = "termcolor"
 version = "1.1.2"
@@ -2149,79 +2200,24 @@ checksum = "360dfd1d6d30e05fda32ace2c8c70e9c0a9da713275777f5a4dbb8a1893930c6"

 [[package]]
 name = "tracing"
-version = "0.1.29"
+version = "0.1.26"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "375a639232caf30edfc78e8d89b2d4c375515393e7af7e16f01cd96917fb2105"
+checksum = "09adeb8c97449311ccd28a427f96fb563e7fd31aabf994189879d9da2394b89d"
 dependencies = [
 "cfg-if 1.0.0",
 "pin-project-lite",
- "tracing-attributes",
 "tracing-core",
 ]

-[[package]]
-name = "tracing-attributes"
-version = "0.1.18"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f4f480b8f81512e825f337ad51e94c1eb5d3bbdf2b363dcd01e2b19a9ffe3f8e"
-dependencies = [
- "proc-macro2",
- "quote",
- "syn",
-]
-
 [[package]]
 name = "tracing-core"
-version = "0.1.21"
+version = "0.1.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1f4ed65637b8390770814083d20756f87bfa2c21bf2f110babdc5438351746e4"
+checksum = "a9ff14f98b1a4b289c6248a023c1c2fa1491062964e9fed67ab29c4e4da4a052"
 dependencies = [
 "lazy_static",
 ]

-[[package]]
-name = "tracing-log"
-version = "0.1.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a6923477a48e41c1951f1999ef8bb5a3023eb723ceadafe78ffb65dc366761e3"
-dependencies = [
- "lazy_static",
- "log",
- "tracing-core",
-]
-
-[[package]]
-name = "tracing-serde"
-version = "0.1.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fb65ea441fbb84f9f6748fd496cf7f63ec9af5bca94dd86456978d055e8eb28b"
-dependencies = [
- "serde",
- "tracing-core",
-]
-
-[[package]]
-name = "tracing-subscriber"
-version = "0.2.25"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0e0d2eaa99c3c2e41547cfa109e910a68ea03823cccad4a0525dcbc9b01e8c71"
-dependencies = [
- "ansi_term 0.12.1",
- "chrono",
- "lazy_static",
- "matchers",
- "regex",
- "serde",
- "serde_json",
- "sharded-slab",
- "smallvec",
- "thread_local",
- "tracing",
- "tracing-core",
- "tracing-log",
- "tracing-serde",
-]
-
 [[package]]
 name = "try-lock"
 version = "0.2.3"
@@ -2320,13 +2316,11 @@ dependencies = [
 "byteorder",
 "bytes",
 "clap",
- "const_format",
 "crc32c",
 "daemonize",
 "fs2",
 "hex",
 "humantime",
- "hyper",
 "lazy_static",
 "log",
 "pageserver",
@@ -2334,15 +2328,12 @@ dependencies = [
 "postgres-protocol",
 "postgres_ffi",
 "regex",
- "routerify",
 "rust-s3",
 "serde",
- "serde_json",
 "tokio",
 "tokio-stream",
 "walkdir",
 "workspace_hack",
- "zenith_metrics",
 "zenith_utils",
 ]

@@ -2563,7 +2554,6 @@ version = "0.1.0"
 dependencies = [
 "lazy_static",
 "libc",
- "once_cell",
 "prometheus",
 ]

@@ -2588,12 +2578,13 @@ dependencies = [
 "rustls-split",
 "serde",
 "serde_json",
- "tempfile",
+ "slog",
+ "slog-async",
+ "slog-scope",
+ "slog-stdlog",
+ "slog-term",
 "thiserror",
 "tokio",
- "tracing",
- "tracing-log",
- "tracing-subscriber",
 "webpki",
 "workspace_hack",
 "zenith_metrics",
--- a/38
+++ b/38
@@ -10,21 +10,39 @@ FROM zenithdb/build:buster AS pg-build
 WORKDIR /zenith
 COPY ./vendor/postgres vendor/postgres
 COPY ./Makefile Makefile
-ENV BUILD_TYPE release
 RUN make -j $(getconf _NPROCESSORS_ONLN) -s postgres
-RUN rm -rf postgres_install/build
+
+#
+# Calculate cargo dependencies.
+# This will always run, but only generate recipe.json with list of dependencies without
+# installing them.
+#
+FROM zenithdb/build:buster AS cargo-deps-inspect
+WORKDIR /zenith
+COPY . .
+RUN cargo chef prepare --recipe-path /zenith/recipe.json
+
+#
+# Build cargo dependencies.
+# This temp cantainner should be rebuilt only if recipe.json was changed.
+#
+FROM zenithdb/build:buster AS deps-build
+WORKDIR /zenith
+COPY --from=pg-build /zenith/tmp_install/include/postgresql/server tmp_install/include/postgresql/server
+COPY --from=cargo-deps-inspect /usr/local/cargo/bin/cargo-chef /usr/local/cargo/bin/
+COPY --from=cargo-deps-inspect /zenith/recipe.json recipe.json
+RUN ROCKSDB_LIB_DIR=/usr/lib/ cargo chef cook --release --recipe-path recipe.json

 #
 # Build zenith binaries
 #
-# TODO: build cargo deps as separate layer. We used cargo-chef before but that was
-# net time waste in a lot of cases. Copying Cargo.lock with empty lib.rs should do the work.
-#
 FROM zenithdb/build:buster AS build
 WORKDIR /zenith
-COPY --from=pg-build /zenith/tmp_install/include/postgresql/server tmp_install/include/postgresql/server
-
 COPY . .
+# Copy cached dependencies
+COPY --from=pg-build /zenith/tmp_install/include/postgresql/server tmp_install/include/postgresql/server
+COPY --from=deps-build /zenith/target target
+COPY --from=deps-build /usr/local/cargo/ /usr/local/cargo/
 RUN cargo build --release

 #
@@ -33,13 +51,11 @@ RUN cargo build --release
 FROM debian:buster-slim
 WORKDIR /data

-RUN apt-get update && apt-get -yq install libreadline-dev libseccomp-dev openssl ca-certificates && \
+RUN apt-get update && apt-get -yq install librocksdb-dev libseccomp-dev openssl && \
    mkdir zenith_install

 COPY --from=build /zenith/target/release/pageserver /usr/local/bin
-COPY --from=build /zenith/target/release/safekeeper /usr/local/bin
-# TODO: temporary alias for compatibility, see https://github.com/zenithdb/zenith/pull/740
-RUN ln -s /usr/local/bin/safekeeper /usr/local/bin/wal_acceptor
+COPY --from=build /zenith/target/release/wal_acceptor /usr/local/bin
 COPY --from=build /zenith/target/release/proxy /usr/local/bin
 COPY --from=pg-build /zenith/tmp_install postgres_install
 COPY docker-entrypoint.sh /docker-entrypoint.sh
--- a/Dockerfile.alpine
+++ b/Dockerfile.alpine
@@ -81,9 +81,7 @@ FROM alpine:3.13
 RUN apk add --update openssl build-base libseccomp-dev
 RUN apk --no-cache --update --repository https://dl-cdn.alpinelinux.org/alpine/edge/testing add rocksdb
 COPY --from=build /zenith/target/release/pageserver /usr/local/bin
-COPY --from=build /zenith/target/release/safekeeper /usr/local/bin
-# TODO: temporary alias for compatibility, see https://github.com/zenithdb/zenith/pull/740
-RUN ln -s /usr/local/bin/safekeeper /usr/local/bin/wal_acceptor
+COPY --from=build /zenith/target/release/wal_acceptor /usr/local/bin
 COPY --from=build /zenith/target/release/proxy /usr/local/bin
 COPY --from=pg-build /zenith/tmp_install /usr/local
 COPY docker-entrypoint.sh /docker-entrypoint.sh
--- a/Dockerfile.build
+++ b/Dockerfile.build
@@ -9,7 +9,7 @@ WORKDIR /zenith
 # Install postgres and zenith build dependencies
 # clang is for rocksdb
 RUN apt-get update && apt-get -yq install automake libtool build-essential bison flex libreadline-dev zlib1g-dev libxml2-dev \
-                                          libseccomp-dev pkg-config libssl-dev clang
+                                          libseccomp-dev pkg-config libssl-dev librocksdb-dev clang

 # Install rust tools
-RUN rustup component add clippy && cargo install cargo-audit
+RUN rustup component add clippy && cargo install cargo-chef cargo-audit
--- a/58
+++ b/58
@@ -6,55 +6,34 @@ else
 	SECCOMP =
 endif

-#
-# We differentiate between release / debug build types using the BUILD_TYPE
-# environment variable.
-#
-BUILD_TYPE ?= debug
-ifeq ($(BUILD_TYPE),release)
-	PG_CONFIGURE_OPTS = --enable-debug
-	PG_CFLAGS = -O2 -g3 $(CFLAGS)
-	# Unfortunately, `--profile=...` is a nightly feature
-	CARGO_BUILD_FLAGS += --release
-else ifeq ($(BUILD_TYPE),debug)
-	PG_CONFIGURE_OPTS = --enable-debug --enable-cassert --enable-depend
-	PG_CFLAGS = -O0 -g3 $(CFLAGS)
-else
-$(error Bad build type `$(BUILD_TYPE)', see Makefile for options)
-endif
-
-# Choose whether we should be silent or verbose
-CARGO_BUILD_FLAGS += --$(if $(filter s,$(MAKEFLAGS)),quiet,verbose)
-# Fix for a corner case when make doesn't pass a jobserver
-CARGO_BUILD_FLAGS += $(filter -j1,$(MAKEFLAGS))
-
-# This option has a side effect of passing make jobserver to cargo.
-# However, we shouldn't do this if `make -n` (--dry-run) has been asked.
-CARGO_CMD_PREFIX += $(if $(filter n,$(MAKEFLAGS)),,+)
-# Force cargo not to print progress bar
-CARGO_CMD_PREFIX += CARGO_TERM_PROGRESS_WHEN=never CI=1
-
 #
 # Top level Makefile to build Zenith and PostgreSQL
 #
-.PHONY: all
 all: zenith postgres

+# We don't want to run 'cargo build' in parallel with the postgres build,
+# because interleaving cargo build output with postgres build output looks
+# confusing. Also, 'cargo build' is parallel on its own, so it would be too
+# much parallelism. (Recursive invocation of postgres target still gets any
+# '-j' flag from the command line, so 'make -j' is still useful.)
+.NOTPARALLEL:
+
 ### Zenith Rust bits
 #
 # The 'postgres_ffi' depends on the Postgres headers.
 .PHONY: zenith
 zenith: postgres-headers
-	+@echo "Compiling Zenith"
-	$(CARGO_CMD_PREFIX) cargo build $(CARGO_BUILD_FLAGS)
+	cargo build

 ### PostgreSQL parts
 tmp_install/build/config.status:
 	+@echo "Configuring postgres build"
 	mkdir -p tmp_install/build
 	(cd tmp_install/build && \
-	../../vendor/postgres/configure CFLAGS='$(PG_CFLAGS)' \
-		$(PG_CONFIGURE_OPTS) \
+	../../vendor/postgres/configure CFLAGS='-O0 -g3 $(CFLAGS)' \
+		--enable-cassert \
+		--enable-debug \
+		--enable-depend \
 		$(SECCOMP) \
 		--prefix=$(abspath tmp_install) > configure.log)

@@ -68,10 +47,10 @@ postgres-headers: postgres-configure
 	+@echo "Installing PostgreSQL headers"
 	$(MAKE) -C tmp_install/build/src/include MAKELEVEL=0 install

+
 # Compile and install PostgreSQL and contrib/zenith
 .PHONY: postgres
-postgres: postgres-configure \
-		  postgres-headers # to prevent `make install` conflicts with zenith's `postgres-headers`
+postgres: postgres-configure
 	+@echo "Compiling PostgreSQL"
 	$(MAKE) -C tmp_install/build MAKELEVEL=0 install
 	+@echo "Compiling contrib/zenith"
@@ -79,21 +58,18 @@ postgres: postgres-configure \
 	+@echo "Compiling contrib/zenith_test_utils"
 	$(MAKE) -C tmp_install/build/contrib/zenith_test_utils install

-.PHONY: postgres-clean
 postgres-clean:
 	$(MAKE) -C tmp_install/build MAKELEVEL=0 clean

 # This doesn't remove the effects of 'configure'.
-.PHONY: clean
 clean:
-	cd tmp_install/build && $(MAKE) clean
-	$(CARGO_CMD_PREFIX) cargo clean
+	cd tmp_install/build && ${MAKE} clean
+	cargo clean

 # This removes everything
-.PHONY: distclean
 distclean:
 	rm -rf tmp_install
-	$(CARGO_CMD_PREFIX) cargo clean
+	cargo clean

 .PHONY: fmt
 fmt:
--- a/README.md
+++ b/README.md
@@ -25,7 +25,7 @@ Pageserver consists of:
 On Ubuntu or Debian this set of packages should be sufficient to build the code:
 ```text
 apt install build-essential libtool libreadline-dev zlib1g-dev flex bison libseccomp-dev \
-libssl-dev clang pkg-config libpq-dev
+libssl-dev clang
 ```

 [Rust] 1.52 or later is also required.
@@ -108,13 +108,6 @@ postgres=# insert into t values(2,2);
 INSERT 0 1
 ```

-6. If you want to run tests afterwards (see below), you have to stop pageserver and all postgres instances you have just started:
-```sh
-> ./target/debug/zenith pg stop migration_check
-> ./target/debug/zenith pg stop main
-> ./target/debug/zenith stop
-```
-
 ## Running tests

 ```sh
--- a/control_plane/src/compute.rs
+++ b/control_plane/src/compute.rs
@@ -1,16 +1,17 @@
-use std::collections::BTreeMap;
-use std::fs::{self, File};
+use std::fs::{self, File, OpenOptions};
 use std::io::Write;
 use std::net::SocketAddr;
 use std::net::TcpStream;
 use std::os::unix::fs::PermissionsExt;
-use std::path::PathBuf;
 use std::process::{Command, Stdio};
 use std::str::FromStr;
 use std::sync::Arc;
 use std::time::Duration;
+use std::{collections::BTreeMap, path::PathBuf};

 use anyhow::{Context, Result};
+use lazy_static::lazy_static;
+use regex::Regex;
 use zenith_utils::connstring::connection_host_port;
 use zenith_utils::lsn::Lsn;
 use zenith_utils::postgres_backend::AuthType;
@@ -18,7 +19,6 @@ use zenith_utils::zid::ZTenantId;
 use zenith_utils::zid::ZTimelineId;

 use crate::local_env::LocalEnv;
-use crate::postgresql_conf::PostgresConf;
 use crate::storage::PageServerNode;

 //
@@ -144,25 +144,76 @@ impl PostgresNode {
            );
        }

+        lazy_static! {
+            static ref CONF_PORT_RE: Regex = Regex::new(r"(?m)^\s*port\s*=\s*(\d+)\s*$").unwrap();
+            static ref CONF_TIMELINE_RE: Regex =
+                Regex::new(r"(?m)^\s*zenith.zenith_timeline\s*=\s*'(\w+)'\s*$").unwrap();
+            static ref CONF_TENANT_RE: Regex =
+                Regex::new(r"(?m)^\s*zenith.zenith_tenant\s*=\s*'(\w+)'\s*$").unwrap();
+        }
+
        // parse data directory name
        let fname = entry.file_name();
        let name = fname.to_str().unwrap().to_string();

-        // Read config file into memory
+        // find out tcp port in config file
        let cfg_path = entry.path().join("postgresql.conf");
-        let cfg_path_str = cfg_path.to_string_lossy();
-        let mut conf_file = File::open(&cfg_path)
-            .with_context(|| format!("failed to open config file in {}", cfg_path_str))?;
-        let conf = PostgresConf::read(&mut conf_file)
-            .with_context(|| format!("failed to read config file in {}", cfg_path_str))?;
+        let config = fs::read_to_string(cfg_path.clone()).with_context(|| {
+            format!(
+                "failed to read config file in {}",
+                cfg_path.to_str().unwrap()
+            )
+        })?;

-        // Read a few options from the config file
-        let context = format!("in config file {}", cfg_path_str);
-        let port: u16 = conf.parse_field("port", &context)?;
-        let timelineid: ZTimelineId = conf.parse_field("zenith.zenith_timeline", &context)?;
-        let tenantid: ZTenantId = conf.parse_field("zenith.zenith_tenant", &context)?;
+        // parse port
+        let err_msg = format!(
+            "failed to find port definition in config file {}",
+            cfg_path.to_str().unwrap()
+        );
+        let port: u16 = CONF_PORT_RE
+            .captures(config.as_str())
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 1"))?
+            .iter()
+            .last()
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 2"))?
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 3"))?
+            .as_str()
+            .parse()
+            .with_context(|| err_msg)?;

-        let uses_wal_proposer = conf.get("wal_acceptors").is_some();
+        // parse timeline
+        let err_msg = format!(
+            "failed to find timeline definition in config file {}",
+            cfg_path.to_str().unwrap()
+        );
+        let timelineid: ZTimelineId = CONF_TIMELINE_RE
+            .captures(config.as_str())
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 1"))?
+            .iter()
+            .last()
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 2"))?
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 3"))?
+            .as_str()
+            .parse()
+            .with_context(|| err_msg)?;
+
+        // parse tenant
+        let err_msg = format!(
+            "failed to find tenant definition in config file {}",
+            cfg_path.to_str().unwrap()
+        );
+        let tenantid = CONF_TENANT_RE
+            .captures(config.as_str())
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 1"))?
+            .iter()
+            .last()
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 2"))?
+            .ok_or_else(|| anyhow::Error::msg(err_msg.clone() + " 3"))?
+            .as_str()
+            .parse()
+            .with_context(|| err_msg)?;
+
+        let uses_wal_proposer = config.contains("wal_acceptors");

        // ok now
        Ok(PostgresNode {
@@ -257,58 +308,74 @@ impl PostgresNode {
    // Connect to a page server, get base backup, and untar it to initialize a
    // new data directory
    fn setup_pg_conf(&self, auth_type: AuthType) -> Result<()> {
-        let mut conf = PostgresConf::new();
-        conf.append("max_wal_senders", "10");
+        File::create(self.pgdata().join("postgresql.conf").to_str().unwrap())?;
+
        // wal_log_hints is mandatory when running against pageserver (see gh issue#192)
        // TODO: is it possible to check wal_log_hints at pageserver side via XLOG_PARAMETER_CHANGE?
-        conf.append("wal_log_hints", "on");
-        conf.append("max_replication_slots", "10");
-        conf.append("hot_standby", "on");
-        conf.append("shared_buffers", "1MB");
-        conf.append("fsync", "off");
-        conf.append("max_connections", "100");
-        conf.append("wal_sender_timeout", "0");
-        conf.append("wal_level", "replica");
-        conf.append("listen_addresses", &self.address.ip().to_string());
-        conf.append("port", &self.address.port().to_string());
+        self.append_conf(
+            "postgresql.conf",
+            &format!(
+                "max_wal_senders = 10\n\
+                 wal_log_hints = on\n\
+                 max_replication_slots = 10\n\
+                 hot_standby = on\n\
+                 shared_buffers = 1MB\n\
+                 fsync = off\n\
+                 max_connections = 100\n\
+                 wal_sender_timeout = 0\n\
+                 wal_level = replica\n\
+                 zenith.file_cache_size = 4096\n\
+                 zenith.file_cache_path = '/tmp/file.cache'\n\
+                 listen_addresses = '{address}'\n\
+                 port = {port}\n",
+                address = self.address.ip(),
+                port = self.address.port()
+            ),
+        )?;

        // Never clean up old WAL. TODO: We should use a replication
        // slot or something proper, to prevent the compute node
        // from removing WAL that hasn't been streamed to the safekeeper or
        // page server yet. (gh issue #349)
-        conf.append("wal_keep_size", "10TB");
+        self.append_conf("postgresql.conf", "wal_keep_size='10TB'\n")?;

-        // Configure the node to fetch pages from pageserver
-        let pageserver_connstr = {
-            let (host, port) = connection_host_port(&self.pageserver.pg_connection_config);
-
-            // Set up authentication
-            //
-            // $ZENITH_AUTH_TOKEN will be replaced with value from environment
-            // variable during compute pg startup. It is done this way because
-            // otherwise user will be able to retrieve the value using SHOW
-            // command or pg_settings
-            let password = if let AuthType::ZenithJWT = auth_type {
-                "$ZENITH_AUTH_TOKEN"
-            } else {
-                ""
-            };
-
-            format!("host={} port={} password={}", host, port, password)
+        // set up authentication
+        let password = if let AuthType::ZenithJWT = auth_type {
+            "$ZENITH_AUTH_TOKEN"
+        } else {
+            ""
        };
-        conf.append("shared_preload_libraries", "zenith");
-        conf.append_line("");
-        conf.append("zenith.page_server_connstring", &pageserver_connstr);
-        conf.append("zenith.zenith_tenant", &self.tenantid.to_string());
-        conf.append("zenith.zenith_timeline", &self.timelineid.to_string());
-        conf.append_line("");
+
+        // Configure that node to take pages from pageserver
+        let (host, port) = connection_host_port(&self.pageserver.pg_connection_config);
+        self.append_conf(
+            "postgresql.conf",
+            format!(
+                concat!(
+                    "shared_preload_libraries = zenith\n",
+                    // $ZENITH_AUTH_TOKEN will be replaced with value from environment variable during compute pg startup
+                    // it is done this way because otherwise user will be able to retrieve the value using SHOW command or pg_settings
+                    "zenith.page_server_connstring = 'host={} port={} password={}'\n",
+                    "zenith.zenith_timeline='{}'\n",
+                    "zenith.zenith_tenant='{}'\n",
+                ),
+                host, port, password, self.timelineid, self.tenantid,
+            )
+            .as_str(),
+        )?;

        // Configure the node to stream WAL directly to the pageserver
-        conf.append("synchronous_standby_names", "pageserver"); // TODO: add a new function arg?
-        conf.append("zenith.callmemaybe_connstring", &self.connstr());
-
-        let mut file = File::create(self.pgdata().join("postgresql.conf"))?;
-        file.write_all(conf.to_string().as_bytes())?;
+        self.append_conf(
+            "postgresql.conf",
+            format!(
+                concat!(
+                    "synchronous_standby_names = 'pageserver'\n", // TODO: add a new function arg?
+                    "zenith.callmemaybe_connstring = '{}'\n",     // FIXME escaping
+                ),
+                self.connstr(),
+            )
+            .as_str(),
+        )?;

        Ok(())
    }
@@ -351,6 +418,14 @@ impl PostgresNode {
        }
    }

+    pub fn append_conf(&self, config: &str, opts: &str) -> Result<()> {
+        OpenOptions::new()
+            .append(true)
+            .open(self.pgdata().join(config).to_str().unwrap())?
+            .write_all(opts.as_bytes())?;
+        Ok(())
+    }
+
    fn pg_ctl(&self, args: &[&str], auth_token: &Option<String>) -> Result<()> {
        let pg_ctl_path = self.env.pg_bin_dir().join("pg_ctl");
        let mut cmd = Command::new(pg_ctl_path);
@@ -452,7 +527,9 @@ impl PostgresNode {
            .output()
            .expect("failed to execute whoami");

-        assert!(output.status.success(), "whoami failed");
+        if !output.status.success() {
+            panic!("whoami failed");
+        }

        String::from_utf8(output.stdout).unwrap().trim().to_string()
    }
--- a/control_plane/src/lib.rs
+++ b/control_plane/src/lib.rs
@@ -12,7 +12,6 @@ use std::path::Path;

 pub mod compute;
 pub mod local_env;
-pub mod postgresql_conf;
 pub mod storage;

 /// Read a PID file
--- a/control_plane/src/postgresql_conf.rs
+++ b/control_plane/src/postgresql_conf.rs
@@ -1,212 +0,0 @@
-///
-/// Module for parsing postgresql.conf file.
-///
-/// NOTE: This doesn't implement the full, correct postgresql.conf syntax. Just
-/// enough to extract a few settings we need in Zenith, assuming you don't do
-/// funny stuff like include-directives or funny escaping.
-use anyhow::{anyhow, bail, Context, Result};
-use lazy_static::lazy_static;
-use regex::Regex;
-use std::collections::HashMap;
-use std::fmt;
-use std::io::BufRead;
-use std::str::FromStr;
-
-/// In-memory representation of a postgresql.conf file
-#[derive(Default)]
-pub struct PostgresConf {
-    lines: Vec<String>,
-    hash: HashMap<String, String>,
-}
-
-lazy_static! {
-    static ref CONF_LINE_RE: Regex = Regex::new(r"^((?:\w|\.)+)\s*=\s*(\S+)$").unwrap();
-}
-
-impl PostgresConf {
-    pub fn new() -> PostgresConf {
-        PostgresConf::default()
-    }
-
-    /// Read file into memory
-    pub fn read(read: impl std::io::Read) -> Result<PostgresConf> {
-        let mut result = Self::new();
-
-        for line in std::io::BufReader::new(read).lines() {
-            let line = line?;
-
-            // Store each line in a vector, in original format
-            result.lines.push(line.clone());
-
-            // Also parse each line and insert key=value lines into a hash map.
-            //
-            // FIXME: This doesn't match exactly the flex/bison grammar in PostgreSQL.
-            // But it's close enough for our usage.
-            let line = line.trim();
-            if line.starts_with('#') {
-                // comment, ignore
-                continue;
-            } else if let Some(caps) = CONF_LINE_RE.captures(line) {
-                let name = caps.get(1).unwrap().as_str();
-                let raw_val = caps.get(2).unwrap().as_str();
-
-                if let Ok(val) = deescape_str(raw_val) {
-                    // Note: if there's already an entry in the hash map for
-                    // this key, this will replace it. That's the behavior what
-                    // we want; when PostgreSQL reads the file, each line
-                    // overrides any previous value for the same setting.
-                    result.hash.insert(name.to_string(), val.to_string());
-                }
-            }
-        }
-        Ok(result)
-    }
-
-    /// Return the current value of 'option'
-    pub fn get(&self, option: &str) -> Option<&str> {
-        self.hash.get(option).map(|x| x.as_ref())
-    }
-
-    /// Return the current value of a field, parsed to the right datatype.
-    ///
-    /// This calls the FromStr::parse() function on the value of the field. If
-    /// the field does not exist, or parsing fails, returns an error.
-    ///
-    pub fn parse_field<T>(&self, field_name: &str, context: &str) -> Result<T>
-    where
-        T: FromStr,
-        <T as FromStr>::Err: std::error::Error + Send + Sync + 'static,
-    {
-        self.get(field_name)
-            .ok_or_else(|| anyhow!("could not find '{}' option {}", field_name, context))?
-            .parse::<T>()
-            .with_context(|| format!("could not parse '{}' option {}", field_name, context))
-    }
-
-    ///
-    /// Note: if you call this multiple times for the same option, the config
-    /// file will a line for each call. It would be nice to have a function
-    /// to change an existing line, but that's a TODO.
-    ///
-    pub fn append(&mut self, option: &str, value: &str) {
-        self.lines
-            .push(format!("{}={}\n", option, escape_str(value)));
-        self.hash.insert(option.to_string(), value.to_string());
-    }
-
-    /// Append an arbitrary non-setting line to the config file
-    pub fn append_line(&mut self, line: &str) {
-        self.lines.push(line.to_string());
-    }
-}
-
-impl fmt::Display for PostgresConf {
-    /// Return the whole configuration file as a string
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        for line in self.lines.iter() {
-            f.write_str(line)?;
-        }
-        Ok(())
-    }
-}
-
-/// Escape a value for putting in postgresql.conf.
-fn escape_str(s: &str) -> String {
-    // If the string doesn't contain anything that needs quoting or escaping, return it
-    // as it is.
-    //
-    // The first part of the regex, before the '|', matches the INTEGER rule in the
-    // PostgreSQL flex grammar (guc-file.l). It matches plain integers like "123" and
-    // "-123", and also accepts units like "10MB". The second part of the regex matches
-    // the UNQUOTED_STRING rule, and accepts strings that contain a single word, beginning
-    // with a letter. That covers words like "off" or "posix". Everything else is quoted.
-    //
-    // This regex is a bit more conservative than the rules in guc-file.l, so we quote some
-    // strings that PostgreSQL would accept without quoting, but that's OK.
-    lazy_static! {
-        static ref UNQUOTED_RE: Regex =
-            Regex::new(r"(^[-+]?[0-9]+[a-zA-Z]*$)|(^[a-zA-Z][a-zA-Z0-9]*$)").unwrap();
-    }
-    if UNQUOTED_RE.is_match(s) {
-        s.to_string()
-    } else {
-        // Otherwise escape and quote it
-        let s = s
-            .replace('\\', "\\\\")
-            .replace('\n', "\\n")
-            .replace('\'', "''");
-
-        "\'".to_owned() + &s + "\'"
-    }
-}
-
-/// De-escape a possibly-quoted value.
-///
-/// See `DeescapeQuotedString` function in PostgreSQL sources for how PostgreSQL
-/// does this.
-fn deescape_str(s: &str) -> Result<String> {
-    // If the string has a quote at the beginning and end, strip them out.
-    if s.len() >= 2 && s.starts_with('\'') && s.ends_with('\'') {
-        let mut result = String::new();
-
-        let mut iter = s[1..(s.len() - 1)].chars().peekable();
-        while let Some(c) = iter.next() {
-            let newc = if c == '\\' {
-                match iter.next() {
-                    Some('b') => '\x08',
-                    Some('f') => '\x0c',
-                    Some('n') => '\n',
-                    Some('r') => '\r',
-                    Some('t') => '\t',
-                    Some('0'..='7') => {
-                        // TODO
-                        bail!("octal escapes not supported");
-                    }
-                    Some(n) => n,
-                    None => break,
-                }
-            } else if c == '\'' && iter.peek() == Some(&'\'') {
-                // doubled quote becomes just one quote
-                iter.next().unwrap()
-            } else {
-                c
-            };
-
-            result.push(newc);
-        }
-        Ok(result)
-    } else {
-        Ok(s.to_string())
-    }
-}
-
-#[test]
-fn test_postgresql_conf_escapes() -> Result<()> {
-    assert_eq!(escape_str("foo bar"), "'foo bar'");
-    // these don't need to be quoted
-    assert_eq!(escape_str("foo"), "foo");
-    assert_eq!(escape_str("123"), "123");
-    assert_eq!(escape_str("+123"), "+123");
-    assert_eq!(escape_str("-10"), "-10");
-    assert_eq!(escape_str("1foo"), "1foo");
-    assert_eq!(escape_str("foo1"), "foo1");
-    assert_eq!(escape_str("10MB"), "10MB");
-    assert_eq!(escape_str("-10kB"), "-10kB");
-
-    // these need quoting and/or escaping
-    assert_eq!(escape_str("foo bar"), "'foo bar'");
-    assert_eq!(escape_str("fo'o"), "'fo''o'");
-    assert_eq!(escape_str("fo\no"), "'fo\\no'");
-    assert_eq!(escape_str("fo\\o"), "'fo\\\\o'");
-    assert_eq!(escape_str("10 cats"), "'10 cats'");
-
-    // Test de-escaping
-    assert_eq!(deescape_str(&escape_str("foo"))?, "foo");
-    assert_eq!(deescape_str(&escape_str("fo'o\nba\\r"))?, "fo'o\nba\\r");
-    assert_eq!(deescape_str("'\\b\\f\\n\\r\\t'")?, "\x08\x0c\n\r\t");
-
-    // octal-escapes are currently not supported
-    assert!(deescape_str("'foo\\7\\07\\007'").is_err());
-
-    Ok(())
-}
--- a/control_plane/src/storage.rs
+++ b/control_plane/src/storage.rs
@@ -85,20 +85,20 @@ impl PageServerNode {
            ),
            env: env.clone(),
            http_client: Client::new(),
-            http_base_url: format!("http://127.0.0.1:{}/v1", env.pageserver_http_port),
+            http_base_url: format!("http://localhost:{}/v1", env.pageserver_http_port),
        }
    }

    fn pageserver_connection_config(password: &str, port: u16) -> Config {
-        format!("postgresql://no_user:{}@127.0.0.1:{}/no_db", password, port)
+        format!("postgresql://no_user:{}@localhost:{}/no_db", password, port)
            .parse()
            .unwrap()
    }

    pub fn init(&self, create_tenant: Option<&str>, enable_auth: bool) -> anyhow::Result<()> {
        let mut cmd = Command::new(self.env.pageserver_bin()?);
-        let listen_pg = format!("127.0.0.1:{}", self.env.pageserver_pg_port);
-        let listen_http = format!("127.0.0.1:{}", self.env.pageserver_http_port);
+        let listen_pg = format!("localhost:{}", self.env.pageserver_pg_port);
+        let listen_http = format!("localhost:{}", self.env.pageserver_http_port);
        let mut args = vec![
            "--init",
            "-D",
@@ -199,45 +199,23 @@ impl PageServerNode {
        bail!("pageserver failed to start in {} seconds", RETRIES);
    }

-    pub fn stop(&self, immediate: bool) -> anyhow::Result<()> {
+    pub fn stop(&self) -> anyhow::Result<()> {
        let pid = read_pidfile(&self.pid_file())?;
        let pid = Pid::from_raw(pid);
-        if immediate {
-            println!("Stop pageserver immediately");
-            if kill(pid, Signal::SIGQUIT).is_err() {
-                bail!("Failed to kill pageserver with pid {}", pid);
-            }
-        } else {
-            println!("Stop pageserver gracefully");
-            if kill(pid, Signal::SIGTERM).is_err() {
-                bail!("Failed to stop pageserver with pid {}", pid);
-            }
+        if kill(pid, Signal::SIGTERM).is_err() {
+            bail!("Failed to kill pageserver with pid {}", pid);
        }

+        // wait for pageserver stop
        let address = connection_address(&self.pg_connection_config);
-
-        // TODO Remove this "timeout" and handle it on caller side instead.
-        // Shutting down may take a long time,
-        // if pageserver checkpoints a lot of data
-        for _ in 0..100 {
-            if let Err(_e) = TcpStream::connect(&address) {
-                println!("Pageserver stopped receiving connections");
-
-                //Now check status
-                match self.check_status() {
-                    Ok(_) => {
-                        println!("Pageserver status is OK. Wait a bit.");
-                        thread::sleep(Duration::from_secs(1));
-                    }
-                    Err(err) => {
-                        println!("Pageserver status is: {}", err);
-                        return Ok(());
-                    }
-                }
-            } else {
-                println!("Pageserver still receives connections");
-                thread::sleep(Duration::from_secs(1));
+        for _ in 0..5 {
+            let stream = TcpStream::connect(&address);
+            thread::sleep(Duration::from_secs(1));
+            if let Err(_e) = stream {
+                println!("Pageserver stopped");
+                return Ok(());
            }
+            println!("Stopping pageserver on {}", address);
        }

        bail!("Failed to stop pageserver with pid {}", pid);
@@ -335,9 +313,8 @@ impl PageServerNode {

 impl Drop for PageServerNode {
    fn drop(&mut self) {
-        // TODO Looks like this flag is never set
        if self.kill_on_exit {
-            let _ = self.stop(true);
+            let _ = self.stop();
        }
    }
 }
--- a/docker-entrypoint.sh
+++ b/docker-entrypoint.sh
@@ -7,7 +7,7 @@ if [ "$1" = 'pageserver' ]; then
        pageserver --init -D /data --postgres-distrib /usr/local
    fi
    echo "Staring pageserver at 0.0.0.0:6400"
-    pageserver -l 0.0.0.0:6400 --listen-http 0.0.0.0:9898 -D /data
+    pageserver -l 0.0.0.0:6400 -D /data
 else
    "$@"
 fi
--- a/docs/README.md
+++ b/docs/README.md
@@ -10,5 +10,5 @@
 - [pageserver/README](/pageserver/README) — pageserver overview.
 - [postgres_ffi/README](/postgres_ffi/README) — Postgres FFI overview.
 - [test_runner/README.md](/test_runner/README.md) — tests infrastructure overview.
- [walkeeper/README](/walkeeper/README) — WAL service overview.
+- [walkeeper/README](/walkeeper/README.md) — WAL service overview.
 - [core_changes.md](core_changes.md) - Description of Zenith changes in Postgres core
--- a/docs/docker.md
+++ b/docs/docker.md
@@ -4,7 +4,7 @@

 Currently we build two main images:

- [zenithdb/zenith](https://hub.docker.com/repository/docker/zenithdb/zenith) — image with pre-built `pageserver`, `safekeeper` and `proxy` binaries and all the required runtime dependencies. Built from [/Dockerfile](/Dockerfile).
+- [zenithdb/zenith](https://hub.docker.com/repository/docker/zenithdb/zenith) — image with pre-built `pageserver`, `wal_acceptor` and `proxy` binaries and all the required runtime dependencies. Built from [/Dockerfile](/Dockerfile).
 - [zenithdb/compute-node](https://hub.docker.com/repository/docker/zenithdb/compute-node) — compute node image with pre-built Postgres binaries from [zenithdb/postgres](https://github.com/zenithdb/postgres).

 And two intermediate images used either to reduce build time or to deliver some additional binary tools from other repos:
--- a/docs/multitenancy.md
+++ b/docs/multitenancy.md
@@ -56,4 +56,4 @@ Tenant id is passed to postgres via GUC the same way as the timeline. Tenant id

 ### Safety

-For now particular tenant can only appear on a particular pageserver. Set of safekeepers are also pinned to particular (tenantid, timeline) pair so there can only be one writer for particular (tenantid, timeline).
+For now particular tenant can only appear on a particular pageserver. Set of WAL acceptors are also pinned to particular (tenantid, timeline) pair so there can only be one writer for particular (tenantid, timeline).
--- a/pageserver/Cargo.toml
+++ b/pageserver/Cargo.toml
@@ -17,7 +17,7 @@ lazy_static = "1.4.0"
 log = "0.4.14"
 clap = "2.33.0"
 daemonize = "0.4.1"
-tokio = { version = "1.11", features = ["process", "macros", "fs", "rt"] }
+tokio = { version = "1.11", features = ["process", "macros", "fs"] }
 postgres-types = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
 postgres-protocol = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
 postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
@@ -34,14 +34,8 @@ toml = "0.5"
 scopeguard = "1.1.0"
 rust-s3 = { version = "0.27.0-rc4", features = ["no-verify-ssl"] }
 async-trait = "0.1"
-const_format = "0.2.21"
-tracing = "0.1.27"
-signal-hook = {version = "0.3.10", features = ["extended-siginfo"] }

 postgres_ffi = { path = "../postgres_ffi" }
 zenith_metrics = { path = "../zenith_metrics" }
 zenith_utils = { path = "../zenith_utils" }
 workspace_hack = { path = "../workspace_hack" }
-
-[dev-dependencies]
-hex-literal = "0.3"
--- a/pageserver/README
+++ b/pageserver/README
@@ -7,9 +7,8 @@ The Page Server has a few different duties:
 - Replay WAL that's applicable to the chunks that the Page Server maintains
 - Backup to S3

-S3 is the main fault-tolerant storage of all data, as there are no Page Server
-replicas. We use a separate fault-tolerant WAL service to reduce latency. It
-keeps track of WAL records which are not syncted to S3 yet.
+
+

 The Page Server consists of multiple threads that operate on a shared
 repository of page versions:
--- a/pageserver/src/basebackup.rs
+++ b/pageserver/src/basebackup.rs
@@ -31,7 +31,7 @@ use zenith_utils::lsn::Lsn;
 pub struct Basebackup<'a> {
    ar: Builder<&'a mut dyn Write>,
    timeline: &'a Arc<dyn Timeline>,
-    pub lsn: Lsn,
+    lsn: Lsn,
    prev_record_lsn: Lsn,
 }

@@ -97,6 +97,7 @@ impl<'a> Basebackup<'a> {
    pub fn send_tarball(&mut self) -> anyhow::Result<()> {
        // Create pgdata subdirs structure
        for dir in pg_constants::PGDATA_SUBDIRS.iter() {
+            info!("send subdir {:?}", *dir);
            let header = new_tar_header_dir(*dir)?;
            self.ar.append(&header, &mut io::empty())?;
        }
--- a/pageserver/src/bin/pageserver.rs
+++ b/pageserver/src/bin/pageserver.rs
@@ -2,44 +2,28 @@
 // Main entry point for the Page Server executable
 //

+use log::*;
 use pageserver::defaults::*;
 use serde::{Deserialize, Serialize};
 use std::{
    env,
    net::TcpListener,
    path::{Path, PathBuf},
+    process::exit,
    str::FromStr,
    thread,
 };
-use tracing::*;
 use zenith_utils::{auth::JwtAuth, logging, postgres_backend::AuthType};

-use anyhow::{bail, ensure, Context, Result};
-use signal_hook::consts::signal::*;
-use signal_hook::consts::TERM_SIGNALS;
-use signal_hook::flag;
-use signal_hook::iterator::exfiltrator::WithOrigin;
-use signal_hook::iterator::SignalsInfo;
-use std::process::exit;
-use std::sync::atomic::AtomicBool;
-use std::sync::Arc;
-
+use anyhow::{bail, ensure, Result};
 use clap::{App, Arg, ArgMatches};
 use daemonize::Daemonize;

 use pageserver::{
-    branches,
-    defaults::{
-        DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_PG_LISTEN_ADDR,
-        DEFAULT_RELISH_STORAGE_MAX_CONCURRENT_SYNC_LIMITS,
-    },
-    http, page_service, relish_storage, tenant_mgr, PageServerConf, RelishStorageConfig,
-    RelishStorageKind, S3Config, LOG_FILE_NAME,
+    branches, http, page_service, tenant_mgr, PageServerConf, RelishStorageConfig, S3Config,
+    LOG_FILE_NAME,
 };
 use zenith_utils::http::endpoint;
-use zenith_utils::postgres_backend;
-
-use const_format::formatcp;

 /// String arguments that can be declared via CLI or config file
 #[derive(Serialize, Deserialize)]
@@ -55,7 +39,6 @@ struct CfgFileParams {
    auth_type: Option<String>,
    // see https://github.com/alexcrichton/toml-rs/blob/6c162e6562c3e432bf04c82a3d1d789d80761a86/examples/enum_external.rs for enum deserialisation examples
    relish_storage: Option<RelishStorage>,
-    relish_storage_max_concurrent_sync: Option<String>,
 }

 #[derive(Serialize, Deserialize, Clone)]
@@ -106,7 +89,6 @@ impl CfgFileParams {
            auth_validation_public_key_path: get_arg("auth-validation-public-key-path"),
            auth_type: get_arg("auth-type"),
            relish_storage,
-            relish_storage_max_concurrent_sync: get_arg("relish-storage-max-concurrent-sync"),
        }
    }

@@ -126,9 +108,6 @@ impl CfgFileParams {
                .or(other.auth_validation_public_key_path),
            auth_type: self.auth_type.or(other.auth_type),
            relish_storage: self.relish_storage.or(other.relish_storage),
-            relish_storage_max_concurrent_sync: self
-                .relish_storage_max_concurrent_sync
-                .or(other.relish_storage_max_concurrent_sync),
        }
    }

@@ -197,34 +176,25 @@ impl CfgFileParams {
            );
        }

-        let max_concurrent_sync = match self.relish_storage_max_concurrent_sync.as_deref() {
-            Some(relish_storage_max_concurrent_sync) => {
-                relish_storage_max_concurrent_sync.parse()?
-            }
-            None => DEFAULT_RELISH_STORAGE_MAX_CONCURRENT_SYNC_LIMITS,
-        };
-        let relish_storage_config = self.relish_storage.as_ref().map(|storage_params| {
-            let storage = match storage_params.clone() {
-                RelishStorage::Local { local_path } => {
-                    RelishStorageKind::LocalFs(PathBuf::from(local_path))
-                }
-                RelishStorage::AwsS3 {
-                    bucket_name,
-                    bucket_region,
-                    access_key_id,
-                    secret_access_key,
-                } => RelishStorageKind::AwsS3(S3Config {
-                    bucket_name,
-                    bucket_region,
-                    access_key_id,
-                    secret_access_key,
-                }),
-            };
-            RelishStorageConfig {
-                max_concurrent_sync,
-                storage,
-            }
-        });
+        let relish_storage_config =
+            self.relish_storage
+                .as_ref()
+                .map(|storage_params| match storage_params.clone() {
+                    RelishStorage::Local { local_path } => {
+                        RelishStorageConfig::LocalFs(PathBuf::from(local_path))
+                    }
+                    RelishStorage::AwsS3 {
+                        bucket_name,
+                        bucket_region,
+                        access_key_id,
+                        secret_access_key,
+                    } => RelishStorageConfig::AwsS3(S3Config {
+                        bucket_name,
+                        bucket_region,
+                        access_key_id,
+                        secret_access_key,
+                    }),
+                });

        Ok(PageServerConf {
            daemonize: false,
@@ -250,7 +220,6 @@ impl CfgFileParams {
 }

 fn main() -> Result<()> {
-    zenith_metrics::set_common_metrics_prefix("pageserver");
    let arg_matches = App::new("Zenith page server")
        .about("Materializes WAL stream to pages and serves them to the postgres")
        .arg(
@@ -259,14 +228,14 @@ fn main() -> Result<()> {
                .long("listen-pg")
                .alias("listen") // keep some compatibility
                .takes_value(true)
-                .help(formatcp!("listen for incoming page requests on ip:port (default: {DEFAULT_PG_LISTEN_ADDR})")),
+                .help("listen for incoming page requests on ip:port (default: 127.0.0.1:5430)"),
        )
        .arg(
            Arg::with_name("listen-http")
                .long("listen-http")
                .alias("http_endpoint") // keep some compatibility
                .takes_value(true)
-                .help(formatcp!("http endpoint address for metrics and management API calls on ip:port (default: {DEFAULT_HTTP_LISTEN_ADDR})")),
+                .help("http endpoint address for for metrics and management API calls ip:port (default: 127.0.0.1:5430)"),
        )
        .arg(
            Arg::with_name("daemonize")
@@ -374,19 +343,10 @@ fn main() -> Result<()> {
                .takes_value(true)
                .help("Credentials to access the AWS S3 bucket"),
        )
-        .arg(
-            Arg::with_name("relish-storage-max-concurrent-sync")
-                .long("relish-storage-max-concurrent-sync")
-                .takes_value(true)
-                .help("Maximum allowed concurrent synchronisations with storage"),
-        )
        .get_matches();

    let workdir = Path::new(arg_matches.value_of("workdir").unwrap_or(".zenith"));
-    let cfg_file_path = workdir
-        .canonicalize()
-        .with_context(|| format!("Error opening workdir '{}'", workdir.display()))?
-        .join("pageserver.toml");
+    let cfg_file_path = workdir.canonicalize()?.join("pageserver.toml");

    let args_params = CfgFileParams::from_args(&arg_matches);

@@ -398,37 +358,22 @@ fn main() -> Result<()> {
        args_params
    } else {
        // Supplement the CLI arguments with the config file
-        let cfg_file_contents = std::fs::read_to_string(&cfg_file_path)
-            .with_context(|| format!("No pageserver config at '{}'", cfg_file_path.display()))?;
-        let file_params: CfgFileParams = toml::from_str(&cfg_file_contents).with_context(|| {
-            format!(
-                "Failed to read '{}' as pageserver config",
-                cfg_file_path.display()
-            )
-        })?;
+        let cfg_file_contents = std::fs::read_to_string(&cfg_file_path)?;
+        let file_params: CfgFileParams = toml::from_str(&cfg_file_contents)?;
        args_params.or(file_params)
    };

    // Set CWD to workdir for non-daemon modes
-    env::set_current_dir(&workdir).with_context(|| {
-        format!(
-            "Failed to set application's current dir to '{}'",
-            workdir.display()
-        )
-    })?;
+    env::set_current_dir(&workdir)?;

    // Ensure the config is valid, even if just init-ing
-    let mut conf = params.try_into_config().with_context(|| {
-        format!(
-            "Pageserver config at '{}' is not valid",
-            cfg_file_path.display()
-        )
-    })?;
+    let mut conf = params.try_into_config()?;

    conf.daemonize = arg_matches.is_present("daemonize");

    if init && conf.daemonize {
-        bail!("--daemonize cannot be used with --init")
+        eprintln!("--daemonize cannot be used with --init");
+        exit(1);
    }

    // The configuration is all set up now. Turn it into a 'static
@@ -438,37 +383,21 @@ fn main() -> Result<()> {

    // Create repo and exit if init was requested
    if init {
-        branches::init_pageserver(conf, create_tenant).context("Failed to init pageserver")?;
+        branches::init_pageserver(conf, create_tenant)?;
        // write the config file
-        let cfg_file_contents = toml::to_string_pretty(&params)
-            .context("Failed to create pageserver config contents for initialisation")?;
+        let cfg_file_contents = toml::to_string_pretty(&params)?;
        // TODO support enable-auth flag
-        std::fs::write(&cfg_file_path, cfg_file_contents).with_context(|| {
-            format!(
-                "Failed to initialize pageserver config at '{}'",
-                cfg_file_path.display()
-            )
-        })?;
-        Ok(())
-    } else {
-        start_pageserver(conf).context("Failed to start pageserver")
+        std::fs::write(&cfg_file_path, cfg_file_contents)?;
+
+        return Ok(());
    }
+
+    start_pageserver(conf)
 }

 fn start_pageserver(conf: &'static PageServerConf) -> Result<()> {
    // Initialize logger
-    let log_file = logging::init(LOG_FILE_NAME, conf.daemonize)?;
-
-    let term_now = Arc::new(AtomicBool::new(false));
-    for sig in TERM_SIGNALS {
-        // When terminated by a second term signal, exit with exit code 1.
-        // This will do nothing the first time (because term_now is false).
-        flag::register_conditional_shutdown(*sig, 1, Arc::clone(&term_now))?;
-        // But this will "arm" the above for the second time, by setting it to true.
-        // The order of registering these is important, if you put this one first, it will
-        // first arm and then terminate ‒ all in the first round.
-        flag::register(*sig, Arc::clone(&term_now))?;
-    }
+    let (_scope_guard, log_file) = logging::init(LOG_FILE_NAME, conf.daemonize)?;

    // TODO: Check that it looks like a valid repository before going further

@@ -501,20 +430,16 @@ fn start_pageserver(conf: &'static PageServerConf) -> Result<()> {

        match daemonize.start() {
            Ok(_) => info!("Success, daemonized"),
-            Err(err) => error!(%err, "could not daemonize"),
+            Err(e) => error!("Error, {}", e),
        }
    }

-    // keep join handles for spawned threads
-    // don't spawn threads before daemonizing
-    let mut join_handles = Vec::new();
-
-    if let Some(handle) = relish_storage::run_storage_sync_thread(conf)? {
-        join_handles.push(handle);
-    }
    // Initialize tenant manager.
    tenant_mgr::init(conf);

+    // keep join handles for spawned threads
+    let mut join_handles = vec![];
+
    // initialize authentication for incoming connections
    let auth = match &conf.auth_type {
        AuthType::Trust | AuthType::MD5 => None,
@@ -546,42 +471,13 @@ fn start_pageserver(conf: &'static PageServerConf) -> Result<()> {
            page_service::thread_main(conf, auth, pageserver_listener, conf.auth_type)
        })?;

-    for info in SignalsInfo::<WithOrigin>::new(TERM_SIGNALS)?.into_iter() {
-        match info.signal {
-            SIGQUIT => {
-                info!("Got SIGQUIT. Terminate pageserver in immediate shutdown mode");
-                exit(111);
-            }
-            SIGTERM => {
-                info!("Got SIGINT/SIGTERM. Terminate gracefully in fast shutdown mode");
-                // Terminate postgres backends
-                postgres_backend::set_pgbackend_shutdown_requested();
-                // Stop all tenants and flush their data
-                tenant_mgr::shutdown_all_tenants()?;
-                // Wait for pageservice thread to complete the job
-                page_service_thread
-                    .join()
-                    .expect("thread panicked")
-                    .expect("thread exited with an error");
+    join_handles.push(page_service_thread);

-                // Shut down http router
-                endpoint::shutdown();
-
-                // Wait for all threads
-                for handle in join_handles.into_iter() {
-                    handle
-                        .join()
-                        .expect("thread panicked")
-                        .expect("thread exited with an error");
-                }
-                info!("Pageserver shut down successfully completed");
-                exit(0);
-            }
-            _ => {
-                debug!("Unknown signal.");
-            }
-        }
+    for handle in join_handles.into_iter() {
+        handle
+            .join()
+            .expect("thread panicked")
+            .expect("thread exited with an error")
    }
-
    Ok(())
 }
--- a/pageserver/src/branches.rs
+++ b/pageserver/src/branches.rs
@@ -14,12 +14,11 @@ use std::{
    str::FromStr,
    sync::Arc,
 };
-use tracing::*;
+use zenith_utils::zid::{ZTenantId, ZTimelineId};

-use zenith_utils::crashsafe_dir;
+use log::*;
 use zenith_utils::logging;
 use zenith_utils::lsn::Lsn;
-use zenith_utils::zid::{ZTenantId, ZTimelineId};

 use crate::tenant_mgr;
 use crate::walredo::WalRedoManager;
@@ -100,7 +99,7 @@ pub struct PointInTime {
 pub fn init_pageserver(conf: &'static PageServerConf, create_tenant: Option<&str>) -> Result<()> {
    // Initialize logger
    // use true as daemonize parameter because otherwise we pollute zenith cli output with a few pages long output of info messages
-    let _log_file = logging::init(LOG_FILE_NAME, true)?;
+    let (_scope_guard, _log_file) = logging::init(LOG_FILE_NAME, true)?;

    // We don't use the real WAL redo manager, because we don't want to spawn the WAL redo
    // process during repository initialization.
@@ -119,7 +118,7 @@ pub fn init_pageserver(conf: &'static PageServerConf, create_tenant: Option<&str
        println!("initializing tenantid {}", tenantid);
        create_repo(conf, tenantid, dummy_redo_mgr).with_context(|| "failed to create repo")?;
    }
-    crashsafe_dir::create_dir_all(conf.tenants_path())?;
+    fs::create_dir_all(conf.tenants_path())?;

    println!("pageserver init succeeded");
    Ok(())
@@ -136,12 +135,12 @@ pub fn create_repo(
    }

    // top-level dir may exist if we are creating it through CLI
-    crashsafe_dir::create_dir_all(&repo_dir)
+    fs::create_dir_all(&repo_dir)
        .with_context(|| format!("could not create directory {}", repo_dir.display()))?;

-    crashsafe_dir::create_dir(conf.timelines_path(&tenantid))?;
-    crashsafe_dir::create_dir_all(conf.branches_path(&tenantid))?;
-    crashsafe_dir::create_dir_all(conf.tags_path(&tenantid))?;
+    fs::create_dir(conf.timelines_path(&tenantid))?;
+    fs::create_dir_all(conf.branches_path(&tenantid))?;
+    fs::create_dir_all(conf.tags_path(&tenantid))?;

    info!("created directory structure in {}", repo_dir.display());

@@ -151,13 +150,12 @@ pub fn create_repo(
        conf,
        wal_redo_manager,
        tenantid,
-        false,
    ));

    // Load data into pageserver
    // TODO To implement zenith import we need to
    //      move data loading out of create_repo()
-    bootstrap_timeline(conf, tenantid, tli, repo.as_ref())?;
+    bootstrap_timeline(conf, tenantid, tli, &*repo)?;

    Ok(repo)
 }
@@ -176,16 +174,13 @@ fn get_lsn_from_controlfile(path: &Path) -> Result<Lsn> {
 // to get bootstrap data for timeline initialization.
 //
 fn run_initdb(conf: &'static PageServerConf, initdbpath: &Path) -> Result<()> {
-    info!("running initdb in {}... ", initdbpath.display());
+    info!("running initdb... ");

    let initdb_path = conf.pg_bin_dir().join("initdb");
    let initdb_output = Command::new(initdb_path)
        .args(&["-D", initdbpath.to_str().unwrap()])
        .args(&["-U", &conf.superuser])
        .arg("--no-instructions")
-        // This is only used for a temporary installation that is deleted shortly after,
-        // so no need to fsync it
-        .arg("--no-sync")
        .env_clear()
        .env("LD_LIBRARY_PATH", conf.pg_lib_dir().to_str().unwrap())
        .env("DYLD_LIBRARY_PATH", conf.pg_lib_dir().to_str().unwrap())
@@ -198,6 +193,7 @@ fn run_initdb(conf: &'static PageServerConf, initdbpath: &Path) -> Result<()> {
            String::from_utf8_lossy(&initdb_output.stderr)
        );
    }
+    info!("initdb succeeded");

    Ok(())
 }
@@ -212,8 +208,6 @@ fn bootstrap_timeline(
    tli: ZTimelineId,
    repo: &dyn Repository,
 ) -> Result<()> {
-    let _enter = info_span!("bootstrapping", timeline = %tli, tenant = %tenantid).entered();
-
    let initdb_path = conf.tenant_path(&tenantid).join("tmp");

    // Init temporarily repo to get bootstrap data
@@ -222,14 +216,12 @@ fn bootstrap_timeline(

    let lsn = get_lsn_from_controlfile(&pgdata_path)?.align();

+    info!("bootstrap_timeline {:?} at lsn {}", pgdata_path, lsn);
+
    // Import the contents of the data directory at the initial checkpoint
    // LSN, and any WAL after that.
    let timeline = repo.create_empty_timeline(tli)?;
-    restore_local_repo::import_timeline_from_postgres_datadir(
-        &pgdata_path,
-        timeline.writer().as_ref(),
-        lsn,
-    )?;
+    restore_local_repo::import_timeline_from_postgres_datadir(&pgdata_path, &*timeline, lsn)?;
    timeline.checkpoint()?;

    println!(
@@ -421,6 +413,7 @@ fn create_timeline(
    let timelinedir = conf.timeline_path(&timelineid, tenantid);

    fs::create_dir(&timelinedir)?;
+    fs::create_dir(&timelinedir.join("wal"))?;

    if let Some(ancestor) = ancestor {
        let data = format!("{}@{}", ancestor.timelineid, ancestor.lsn);
--- a/pageserver/src/http/routes.rs
+++ b/pageserver/src/http/routes.rs
@@ -1,3 +1,4 @@
+use std::str::FromStr;
 use std::sync::Arc;

 use anyhow::Result;
@@ -5,7 +6,6 @@ use hyper::header;
 use hyper::StatusCode;
 use hyper::{Body, Request, Response, Uri};
 use routerify::{ext::RequestExt, RouterBuilder};
-use tracing::*;
 use zenith_utils::auth::JwtAuth;
 use zenith_utils::http::endpoint::attach_openapi_ui;
 use zenith_utils::http::endpoint::auth_middleware;
@@ -15,8 +15,6 @@ use zenith_utils::http::{
    endpoint,
    error::HttpErrorBody,
    json::{json_request, json_response},
-    request::get_request_param,
-    request::parse_request_param,
 };

 use super::models::BranchCreateRequest;
@@ -58,6 +56,33 @@ fn get_config(request: &Request<Body>) -> &'static PageServerConf {
    get_state(request).conf
 }

+fn get_request_param<'a>(
+    request: &'a Request<Body>,
+    param_name: &str,
+) -> Result<&'a str, ApiError> {
+    match request.param(param_name) {
+        Some(arg) => Ok(arg),
+        None => {
+            return Err(ApiError::BadRequest(format!(
+                "no {} specified in path param",
+                param_name
+            )))
+        }
+    }
+}
+
+fn parse_request_param<T: FromStr>(
+    request: &Request<Body>,
+    param_name: &str,
+) -> Result<T, ApiError> {
+    match get_request_param(request, param_name)?.parse() {
+        Ok(v) => Ok(v),
+        Err(_) => Err(ApiError::BadRequest(
+            "failed to parse tenant id".to_string(),
+        )),
+    }
+}
+
 // healthcheck handler
 async fn status_handler(_: Request<Body>) -> Result<Response<Body>, ApiError> {
    Ok(Response::builder()
@@ -73,7 +98,6 @@ async fn branch_create_handler(mut request: Request<Body>) -> Result<Response<Bo
    check_permission(&request, Some(request_data.tenant_id))?;

    let response_data = tokio::task::spawn_blocking(move || {
-        let _enter = info_span!("/branch_create", name = %request_data.name, tenant = %request_data.tenant_id, startpoint=%request_data.start_point).entered();
        branches::create_branch(
            get_config(&request),
            &request_data.name,
@@ -92,7 +116,6 @@ async fn branch_list_handler(request: Request<Body>) -> Result<Response<Body>, A
    check_permission(&request, Some(tenantid))?;

    let response_data = tokio::task::spawn_blocking(move || {
-        let _enter = info_span!("branch_list", tenant = %tenantid).entered();
        crate::branches::get_branches(get_config(&request), &tenantid)
    })
    .await
@@ -103,12 +126,11 @@ async fn branch_list_handler(request: Request<Body>) -> Result<Response<Body>, A
 // TODO add to swagger
 async fn branch_detail_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
    let tenantid: ZTenantId = parse_request_param(&request, "tenant_id")?;
-    let branch_name: String = get_request_param(&request, "branch_name")?.to_string();
+    let branch_name: &str = get_request_param(&request, "branch_name")?;
    let conf = get_state(&request).conf;
-    let path = conf.branch_path(&branch_name, &tenantid);
+    let path = conf.branch_path(branch_name, &tenantid);

    let response_data = tokio::task::spawn_blocking(move || {
-        let _enter = info_span!("branch_detail", tenant = %tenantid, branch=%branch_name).entered();
        let repo = tenant_mgr::get_repository_for_tenant(tenantid)?;
        BranchInfo::from_path(path, conf, &tenantid, &repo)
    })
@@ -122,13 +144,10 @@ async fn tenant_list_handler(request: Request<Body>) -> Result<Response<Body>, A
    // check for management permission
    check_permission(&request, None)?;

-    let response_data = tokio::task::spawn_blocking(move || {
-        let _enter = info_span!("tenant_list").entered();
-        crate::branches::get_tenants(get_config(&request))
-    })
-    .await
-    .map_err(ApiError::from_err)??;
-
+    let response_data =
+        tokio::task::spawn_blocking(move || crate::branches::get_tenants(get_config(&request)))
+            .await
+            .map_err(ApiError::from_err)??;
    Ok(json_response(StatusCode::OK, response_data)?)
 }

@@ -139,7 +158,6 @@ async fn tenant_create_handler(mut request: Request<Body>) -> Result<Response<Bo
    let request_data: TenantCreateRequest = json_request(&mut request).await?;

    let response_data = tokio::task::spawn_blocking(move || {
-        let _enter = info_span!("tenant_create", tenant = %request_data.tenant_id).entered();
        tenant_mgr::create_repository_for_tenant(get_config(&request), request_data.tenant_id)
    })
    .await
--- a/pageserver/src/layered_repository.rs
+++ b/pageserver/src/layered_repository.rs
--- a/pageserver/src/layered_repository/README.md
+++ b/pageserver/src/layered_repository/README.md
@@ -25,13 +25,11 @@ OnDisk layers can be Image or Delta:
 Dropped segments are always represented on disk by DeltaLayer.

 LSN range defined by start_lsn and end_lsn:
- start_lsn is inclusive.
- end_lsn is exclusive.
-
-For an open in-memory layer, the end_lsn is MAX_LSN. For a frozen
-in-memory layer or a delta layer, it is a valid end bound. An image
-layer represents snapshot at one LSN, so end_lsn is always the
-snapshot LSN + 1
+- start_lsn is always inclusive.
+- end_lsn depends on layer kind:
+	- InMemoryLayer is either unbounded (end_lsn = MAX_LSN) or dropped (end_lsn = drop_lsn)
+    - ImageLayer represents snapshot at one LSN, so end_lsn = lsn.
+    - DeltaLayer has explicit end_lsn, which represents end of incremental layer.

 Layers can be open or historical:
 - Open layer is a writeable one. Only InMemory layer can be open.
--- a/pageserver/src/layered_repository/delta_layer.rs
+++ b/pageserver/src/layered_repository/delta_layer.rs
@@ -42,13 +42,15 @@ use crate::layered_repository::filename::{DeltaFileName, PathOrConf};
 use crate::layered_repository::storage_layer::{
    Layer, PageReconstructData, PageReconstructResult, PageVersion, SegmentTag,
 };
+use crate::repository::WALRecord;
 use crate::waldecoder;
 use crate::PageServerConf;
 use crate::{ZTenantId, ZTimelineId};
-use anyhow::{bail, ensure, Result};
+use anyhow::{bail, Result};
+use bytes::Bytes;
 use log::*;
 use serde::{Deserialize, Serialize};
-use zenith_utils::vec_map::VecMap;
+use std::collections::BTreeMap;
 // avoid binding to Write (conflicts with std::io::Write)
 // while being able to use std::fmt::Write's methods
 use std::fmt::Write as _;
@@ -57,7 +59,7 @@ use std::fs::File;
 use std::io::{BufWriter, Write};
 use std::ops::Bound::Included;
 use std::path::{Path, PathBuf};
-use std::sync::{Mutex, MutexGuard};
+use std::sync::{Arc, Mutex, MutexGuard};

 use bookfile::{Book, BookWriter};

@@ -107,6 +109,12 @@ impl From<&DeltaLayer> for Summary {
    }
 }

+#[derive(Serialize, Deserialize)]
+struct PageVersionMeta {
+    page_image_range: Option<BlobRange>,
+    record_range: Option<BlobRange>,
+}
+
 ///
 /// DeltaLayer is the in-memory data structure associated with an
 /// on-disk delta file.  We keep a DeltaLayer in memory for each
@@ -131,6 +139,9 @@ pub struct DeltaLayer {

    dropped: bool,

+    /// Predecessor layer
+    predecessor: Option<Arc<dyn Layer>>,
+
    inner: Mutex<DeltaLayerInner>,
 }

@@ -141,10 +152,10 @@ pub struct DeltaLayerInner {

    /// All versions of all pages in the file are are kept here.
    /// Indexed by block number and LSN.
-    page_version_metas: VecMap<(u32, Lsn), BlobRange>,
+    page_version_metas: BTreeMap<(u32, Lsn), PageVersionMeta>,

    /// `relsizes` tracks the size of the relation at different points in time.
-    relsizes: VecMap<Lsn, u32>,
+    relsizes: BTreeMap<Lsn, u32>,
 }

 impl Layer for DeltaLayer {
@@ -169,7 +180,29 @@ impl Layer for DeltaLayer {
    }

    fn filename(&self) -> PathBuf {
-        PathBuf::from(self.layer_name().to_string())
+        PathBuf::from(
+            DeltaFileName {
+                seg: self.seg,
+                start_lsn: self.start_lsn,
+                end_lsn: self.end_lsn,
+                dropped: self.dropped,
+            }
+            .to_string(),
+        )
+    }
+
+    fn path(&self) -> Option<PathBuf> {
+        Some(Self::path_for(
+            &self.path_or_conf,
+            self.timelineid,
+            self.tenantid,
+            &DeltaFileName {
+                seg: self.seg,
+                start_lsn: self.start_lsn,
+                end_lsn: self.end_lsn,
+                dropped: self.dropped,
+            },
+        ))
    }

    /// Look up given page in the cache.
@@ -193,22 +226,20 @@ impl Layer for DeltaLayer {
            // Scan the metadata BTreeMap backwards, starting from the given entry.
            let minkey = (blknum, Lsn(0));
            let maxkey = (blknum, lsn);
-            let iter = inner
+            let mut iter = inner
                .page_version_metas
-                .slice_range((Included(&minkey), Included(&maxkey)))
-                .iter()
-                .rev();
-            for ((_blknum, pv_lsn), blob_range) in iter {
-                let pv = PageVersion::des(&read_blob(&page_version_reader, blob_range)?)?;
-
-                if let Some(img) = pv.page_image {
+                .range((Included(&minkey), Included(&maxkey)));
+            while let Some(((_blknum, _entry_lsn), entry)) = iter.next_back() {
+                if let Some(img_range) = &entry.page_image_range {
                    // Found a page image, return it
+                    let img = Bytes::from(read_blob(&page_version_reader, img_range)?);
                    reconstruct_data.page_img = Some(img);
                    need_image = false;
                    break;
-                } else if let Some(rec) = pv.record {
+                } else if let Some(rec_range) = &entry.record_range {
+                    let rec = WALRecord::des(&read_blob(&page_version_reader, rec_range)?)?;
                    let will_init = rec.will_init;
-                    reconstruct_data.records.push((*pv_lsn, rec));
+                    reconstruct_data.records.push(rec);
                    if will_init {
                        // This WAL record initializes the page, so no need to go further back
                        need_image = false;
@@ -224,9 +255,16 @@ impl Layer for DeltaLayer {
        }

        // If an older page image is needed to reconstruct the page, let the
-        // caller know.
+        // caller know about the predecessor layer.
        if need_image {
-            Ok(PageReconstructResult::Continue(self.start_lsn))
+            if let Some(cont_layer) = &self.predecessor {
+                Ok(PageReconstructResult::Continue(
+                    self.start_lsn,
+                    Arc::clone(cont_layer),
+                ))
+            } else {
+                Ok(PageReconstructResult::Missing(self.start_lsn))
+            }
        } else {
            Ok(PageReconstructResult::Complete)
        }
@@ -235,22 +273,21 @@ impl Layer for DeltaLayer {
    /// Get size of the relation at given LSN
    fn get_seg_size(&self, lsn: Lsn) -> Result<u32> {
        assert!(lsn >= self.start_lsn);
-        ensure!(
-            self.seg.rel.is_blocky(),
-            "get_seg_size() called on a non-blocky rel"
-        );

        // Scan the BTreeMap backwards, starting from the given entry.
        let inner = self.load()?;
-        let slice = inner
-            .relsizes
-            .slice_range((Included(&Lsn(0)), Included(&lsn)));
+        let mut iter = inner.relsizes.range((Included(&Lsn(0)), Included(&lsn)));

-        if let Some((_entry_lsn, entry)) = slice.last() {
-            Ok(*entry)
+        let result;
+        if let Some((_entry_lsn, entry)) = iter.next_back() {
+            result = *entry;
+        // Use the base image if needed
+        } else if let Some(predecessor) = &self.predecessor {
+            result = predecessor.get_seg_size(lsn)?;
        } else {
-            Err(anyhow::anyhow!("could not find seg size in delta layer"))
+            result = 0;
        }
+        Ok(result)
    }

    /// Does this segment exist at given LSN?
@@ -270,15 +307,17 @@ impl Layer for DeltaLayer {
    ///
    fn unload(&self) -> Result<()> {
        let mut inner = self.inner.lock().unwrap();
-        inner.page_version_metas = VecMap::default();
-        inner.relsizes = VecMap::default();
+        inner.page_version_metas = BTreeMap::new();
+        inner.relsizes = BTreeMap::new();
        inner.loaded = false;
        Ok(())
    }

    fn delete(&self) -> Result<()> {
        // delete underlying file
-        fs::remove_file(self.path())?;
+        if let Some(path) = self.path() {
+            fs::remove_file(path)?;
+        }
        Ok(())
    }

@@ -295,22 +334,22 @@ impl Layer for DeltaLayer {

        println!("--- relsizes ---");
        let inner = self.load()?;
-        for (k, v) in inner.relsizes.as_slice() {
+        for (k, v) in inner.relsizes.iter() {
            println!("  {}: {}", k, v);
        }
        println!("--- page versions ---");
        let (_path, book) = self.open_book()?;
        let chapter = book.chapter_reader(PAGE_VERSIONS_CHAPTER)?;
-        for ((blk, lsn), blob_range) in inner.page_version_metas.as_slice() {
+        for (k, v) in inner.page_version_metas.iter() {
            let mut desc = String::new();

-            let buf = read_blob(&chapter, blob_range)?;
-            let pv = PageVersion::des(&buf)?;
-
-            if let Some(img) = pv.page_image.as_ref() {
-                write!(&mut desc, " img {} bytes", img.len())?;
+            if let Some(page_image_range) = v.page_image_range.as_ref() {
+                let image = read_blob(&chapter, page_image_range)?;
+                write!(&mut desc, " img {} bytes", image.len())?;
            }
-            if let Some(rec) = pv.record.as_ref() {
+            if let Some(record_range) = v.record_range.as_ref() {
+                let record_bytes = read_blob(&chapter, record_range)?;
+                let rec = WALRecord::des(&record_bytes)?;
                let wal_desc = waldecoder::describe_wal_record(&rec.rec);
                write!(
                    &mut desc,
@@ -320,7 +359,7 @@ impl Layer for DeltaLayer {
                    wal_desc
                )?;
            }
-            println!("  blk {} at {}: {}", blk, lsn, desc);
+            println!("  blk {} at {}: {}", k.0, k.1, desc);
        }

        Ok(())
@@ -342,15 +381,14 @@ impl DeltaLayer {
        }
    }

-    /// Create a new delta file, using the given page versions and relsizes.
-    /// The page versions are passed by an iterator; the iterator must return
-    /// page versions in blknum+lsn order.
+    /// Create a new delta file, using the given btreemaps containing the page versions and
+    /// relsizes.
    ///
    /// This is used to write the in-memory layer to disk. The in-memory layer uses the same
    /// data structure with two btreemaps as we do, so passing the btreemaps is currently
    /// expedient.
    #[allow(clippy::too_many_arguments)]
-    pub fn create<'a>(
+    pub fn create(
        conf: &'static PageServerConf,
        timelineid: ZTimelineId,
        tenantid: ZTenantId,
@@ -358,13 +396,10 @@ impl DeltaLayer {
        start_lsn: Lsn,
        end_lsn: Lsn,
        dropped: bool,
-        page_versions: impl Iterator<Item = (u32, Lsn, &'a PageVersion)>,
-        relsizes: VecMap<Lsn, u32>,
+        predecessor: Option<Arc<dyn Layer>>,
+        page_versions: BTreeMap<(u32, Lsn), PageVersion>,
+        relsizes: BTreeMap<Lsn, u32>,
    ) -> Result<DeltaLayer> {
-        if seg.rel.is_blocky() {
-            assert!(!relsizes.is_empty());
-        }
-
        let delta_layer = DeltaLayer {
            path_or_conf: PathOrConf::Conf(conf),
            timelineid,
@@ -375,14 +410,17 @@ impl DeltaLayer {
            dropped,
            inner: Mutex::new(DeltaLayerInner {
                loaded: true,
-                page_version_metas: VecMap::default(),
+                page_version_metas: BTreeMap::new(),
                relsizes,
            }),
+            predecessor,
        };
        let mut inner = delta_layer.inner.lock().unwrap();

        // Write the in-memory btreemaps into a file
-        let path = delta_layer.path();
+        let path = delta_layer
+            .path()
+            .expect("DeltaLayer is supposed to have a layer path on disk");

        // Note: This overwrites any existing file. There shouldn't be any.
        // FIXME: throw an error instead?
@@ -392,27 +430,42 @@ impl DeltaLayer {

        let mut page_version_writer = BlobWriter::new(book, PAGE_VERSIONS_CHAPTER);

-        for (blknum, lsn, page_version) in page_versions {
-            let buf = PageVersion::ser(page_version)?;
-            let blob_range = page_version_writer.write_blob(&buf)?;
+        for (key, page_version) in page_versions {
+            let page_image_range = page_version
+                .page_image
+                .map(|page_image| page_version_writer.write_blob(page_image.as_ref()))
+                .transpose()?;

-            inner
-                .page_version_metas
-                .append((blknum, lsn), blob_range)
-                .unwrap();
+            let record_range = page_version
+                .record
+                .map(|record| {
+                    let buf = WALRecord::ser(&record)?;
+                    page_version_writer.write_blob(&buf)
+                })
+                .transpose()?;
+
+            let old = inner.page_version_metas.insert(
+                key,
+                PageVersionMeta {
+                    page_image_range,
+                    record_range,
+                },
+            );
+
+            assert!(old.is_none());
        }

        let book = page_version_writer.close()?;

        // Write out page versions
        let mut chapter = book.new_chapter(PAGE_VERSION_METAS_CHAPTER);
-        let buf = VecMap::ser(&inner.page_version_metas)?;
+        let buf = BTreeMap::ser(&inner.page_version_metas)?;
        chapter.write_all(&buf)?;
        let book = chapter.close()?;

        // and relsizes to separate chapter
        let mut chapter = book.new_chapter(REL_SIZES_CHAPTER);
-        let buf = VecMap::ser(&inner.relsizes)?;
+        let buf = BTreeMap::ser(&inner.relsizes)?;
        chapter.write_all(&buf)?;
        let book = chapter.close()?;

@@ -431,8 +484,7 @@ impl DeltaLayer {
        let book = chapter.close()?;

        // This flushes the underlying 'buf_writer'.
-        let writer = book.close()?;
-        writer.get_ref().sync_all()?;
+        book.close()?;

        trace!("saved {}", &path.display());

@@ -446,7 +498,12 @@ impl DeltaLayer {
            &self.path_or_conf,
            self.timelineid,
            self.tenantid,
-            &self.layer_name(),
+            &DeltaFileName {
+                seg: self.seg,
+                start_lsn: self.start_lsn,
+                end_lsn: self.end_lsn,
+                dropped: self.dropped,
+            },
        );

        let file = File::open(&path)?;
@@ -494,10 +551,10 @@ impl DeltaLayer {
        }

        let chapter = book.read_chapter(PAGE_VERSION_METAS_CHAPTER)?;
-        let page_version_metas = VecMap::des(&chapter)?;
+        let page_version_metas = BTreeMap::des(&chapter)?;

        let chapter = book.read_chapter(REL_SIZES_CHAPTER)?;
-        let relsizes = VecMap::des(&chapter)?;
+        let relsizes = BTreeMap::des(&chapter)?;

        debug!("loaded from {}", &path.display());

@@ -516,6 +573,7 @@ impl DeltaLayer {
        timelineid: ZTimelineId,
        tenantid: ZTenantId,
        filename: &DeltaFileName,
+        predecessor: Option<Arc<dyn Layer>>,
    ) -> DeltaLayer {
        DeltaLayer {
            path_or_conf: PathOrConf::Conf(conf),
@@ -527,9 +585,10 @@ impl DeltaLayer {
            dropped: filename.dropped,
            inner: Mutex::new(DeltaLayerInner {
                loaded: false,
-                page_version_metas: VecMap::default(),
-                relsizes: VecMap::default(),
+                page_version_metas: BTreeMap::new(),
+                relsizes: BTreeMap::new(),
            }),
+            predecessor,
        }
    }

@@ -550,28 +609,10 @@ impl DeltaLayer {
            dropped: summary.dropped,
            inner: Mutex::new(DeltaLayerInner {
                loaded: false,
-                page_version_metas: VecMap::default(),
-                relsizes: VecMap::default(),
+                page_version_metas: BTreeMap::new(),
+                relsizes: BTreeMap::new(),
            }),
+            predecessor: None,
        })
    }
-
-    fn layer_name(&self) -> DeltaFileName {
-        DeltaFileName {
-            seg: self.seg,
-            start_lsn: self.start_lsn,
-            end_lsn: self.end_lsn,
-            dropped: self.dropped,
-        }
-    }
-
-    /// Path to the layer file in pageserver workdir.
-    pub fn path(&self) -> PathBuf {
-        Self::path_for(
-            &self.path_or_conf,
-            self.timelineid,
-            self.tenantid,
-            &self.layer_name(),
-        )
-    }
 }
--- a/pageserver/src/layered_repository/filename.rs
+++ b/pageserver/src/layered_repository/filename.rs
@@ -13,8 +13,6 @@ use anyhow::Result;
 use log::*;
 use zenith_utils::lsn::Lsn;

-use super::METADATA_FILE_NAME;
-
 // Note: LayeredTimeline::load_layer_map() relies on this sort order
 #[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone)]
 pub struct DeltaFileName {
@@ -37,7 +35,7 @@ impl DeltaFileName {
    /// Parse a string as a delta file name. Returns None if the filename does not
    /// match the expected pattern.
    ///
-    pub fn parse_str(fname: &str) -> Option<Self> {
+    pub fn from_str(fname: &str) -> Option<Self> {
        let rel;
        let mut parts;
        if let Some(rest) = fname.strip_prefix("rel_") {
@@ -170,7 +168,7 @@ impl ImageFileName {
    /// Parse a string as an image file name. Returns None if the filename does not
    /// match the expected pattern.
    ///
-    pub fn parse_str(fname: &str) -> Option<Self> {
+    pub fn from_str(fname: &str) -> Option<Self> {
        let rel;
        let mut parts;
        if let Some(rest) = fname.strip_prefix("rel_") {
@@ -288,11 +286,11 @@ pub fn list_files(
        let fname = direntry?.file_name();
        let fname = fname.to_str().unwrap();

-        if let Some(deltafilename) = DeltaFileName::parse_str(fname) {
+        if let Some(deltafilename) = DeltaFileName::from_str(fname) {
            deltafiles.push(deltafilename);
-        } else if let Some(imgfilename) = ImageFileName::parse_str(fname) {
+        } else if let Some(imgfilename) = ImageFileName::from_str(fname) {
            imgfiles.push(imgfilename);
-        } else if fname == METADATA_FILE_NAME || fname == "ancestor" || fname.ends_with(".old") {
+        } else if fname == "wal" || fname == "metadata" || fname == "ancestor" {
            // ignore these
        } else {
            warn!("unrecognized filename in timeline dir: {}", fname);
--- a/pageserver/src/layered_repository/image_layer.rs
+++ b/pageserver/src/layered_repository/image_layer.rs
@@ -114,7 +114,25 @@ pub struct ImageLayerInner {

 impl Layer for ImageLayer {
    fn filename(&self) -> PathBuf {
-        PathBuf::from(self.layer_name().to_string())
+        PathBuf::from(
+            ImageFileName {
+                seg: self.seg,
+                lsn: self.lsn,
+            }
+            .to_string(),
+        )
+    }
+
+    fn path(&self) -> Option<PathBuf> {
+        Some(Self::path_for(
+            &self.path_or_conf,
+            self.timelineid,
+            self.tenantid,
+            &ImageFileName {
+                seg: self.seg,
+                lsn: self.lsn,
+            },
+        ))
    }

    fn get_timeline_id(&self) -> ZTimelineId {
@@ -134,8 +152,7 @@ impl Layer for ImageLayer {
    }

    fn get_end_lsn(&self) -> Lsn {
-        // End-bound is exclusive
-        self.lsn + 1
+        self.lsn
    }

    /// Look up given page in the file
@@ -204,7 +221,9 @@ impl Layer for ImageLayer {

    fn delete(&self) -> Result<()> {
        // delete underlying file
-        fs::remove_file(self.path())?;
+        if let Some(path) = self.path() {
+            fs::remove_file(path)?;
+        }
        Ok(())
    }

@@ -280,7 +299,9 @@ impl ImageLayer {
        let inner = layer.inner.lock().unwrap();

        // Write the images into a file
-        let path = layer.path();
+        let path = layer
+            .path()
+            .expect("ImageLayer is supposed to have a layer path on disk");
        // Note: This overwrites any existing file. There shouldn't be any.
        // FIXME: throw an error instead?
        let file = File::create(&path)?;
@@ -315,10 +336,9 @@ impl ImageLayer {
        let book = chapter.close()?;

        // This flushes the underlying 'buf_writer'.
-        let writer = book.close()?;
-        writer.get_ref().sync_all()?;
+        book.close()?;

-        trace!("saved {}", path.display());
+        trace!("saved {}", &path.display());

        drop(inner);

@@ -423,7 +443,15 @@ impl ImageLayer {
    }

    fn open_book(&self) -> Result<(PathBuf, Book<File>)> {
-        let path = self.path();
+        let path = Self::path_for(
+            &self.path_or_conf,
+            self.timelineid,
+            self.tenantid,
+            &ImageFileName {
+                seg: self.seg,
+                lsn: self.lsn,
+            },
+        );

        let file = File::open(&path)?;
        let book = Book::new(file)?;
@@ -470,21 +498,4 @@ impl ImageLayer {
            }),
        })
    }
-
-    fn layer_name(&self) -> ImageFileName {
-        ImageFileName {
-            seg: self.seg,
-            lsn: self.lsn,
-        }
-    }
-
-    /// Path to the layer file in pageserver workdir.
-    pub fn path(&self) -> PathBuf {
-        Self::path_for(
-            &self.path_or_conf,
-            self.timelineid,
-            self.tenantid,
-            &self.layer_name(),
-        )
-    }
 }
--- a/pageserver/src/layered_repository/inmemory_layer.rs
+++ b/pageserver/src/layered_repository/inmemory_layer.rs
@@ -12,17 +12,16 @@ use crate::layered_repository::{DeltaLayer, ImageLayer};
 use crate::repository::WALRecord;
 use crate::PageServerConf;
 use crate::{ZTenantId, ZTimelineId};
-use anyhow::{bail, ensure, Result};
+use anyhow::{bail, Result};
 use bytes::Bytes;
 use log::*;
+use std::collections::BTreeMap;
+use std::ops::Bound::Included;
 use std::path::PathBuf;
-use std::sync::{Arc, RwLock};
-use zenith_utils::vec_map::VecMap;
+use std::sync::{Arc, Mutex};

 use zenith_utils::lsn::Lsn;

-use super::page_versions::PageVersions;
-
 pub struct InMemoryLayer {
    conf: &'static PageServerConf,
    tenantid: ZTenantId,
@@ -35,56 +34,57 @@ pub struct InMemoryLayer {
    ///
    start_lsn: Lsn,

+    /// Frozen in-memory layers have an inclusive end LSN.
+    end_lsn: Option<Lsn>,
+
    /// LSN of the oldest page version stored in this layer
    oldest_pending_lsn: Lsn,

    /// The above fields never change. The parts that do change are in 'inner',
    /// and protected by mutex.
-    inner: RwLock<InMemoryLayerInner>,
-
-    /// Predecessor layer might be needed?
-    incremental: bool,
+    inner: Mutex<InMemoryLayerInner>,
 }

 pub struct InMemoryLayerInner {
-    /// Frozen in-memory layers have an exclusive end LSN.
-    /// Writes are only allowed when this is None
-    end_lsn: Option<Lsn>,
-
    /// If this relation was dropped, remember when that happened.
-    /// The drop LSN is recorded in [`end_lsn`].
-    dropped: bool,
+    drop_lsn: Option<Lsn>,

    ///
    /// All versions of all pages in the layer are are kept here.
    /// Indexed by block number and LSN.
    ///
-    page_versions: PageVersions,
+    page_versions: BTreeMap<(u32, Lsn), PageVersion>,

    ///
    /// `segsizes` tracks the size of the segment at different points in time.
    ///
-    /// For a blocky rel, there is always one entry, at the layer's start_lsn,
-    /// so that determining the size never depends on the predecessor layer. For
-    /// a non-blocky rel, 'segsizes' is not used and is always empty.
-    ///
-    segsizes: VecMap<Lsn, u32>,
+    segsizes: BTreeMap<Lsn, u32>,
+
+    /// Writes are only allowed when true.
+    /// Set to false when this layer is in the process of being replaced.
+    writeable: bool,
+
+    /// Predecessor layer
+    predecessor: Option<Arc<dyn Layer>>,
 }

 impl InMemoryLayerInner {
-    fn assert_writeable(&self) {
-        assert!(self.end_lsn.is_none());
+    fn check_writeable(&self) -> WriteResult<()> {
+        if self.writeable {
+            Ok(())
+        } else {
+            Err(NonWriteableError)
+        }
    }

    fn get_seg_size(&self, lsn: Lsn) -> u32 {
        // Scan the BTreeMap backwards, starting from the given entry.
-        let slice = self.segsizes.slice_range(..=lsn);
+        let mut iter = self.segsizes.range((Included(&Lsn(0)), Included(&lsn)));

-        // We make sure there is always at least one entry
-        if let Some((_entry_lsn, entry)) = slice.last() {
+        if let Some((_entry_lsn, entry)) = iter.next_back() {
            *entry
        } else {
-            panic!("could not find seg size in in-memory layer");
+            0
        }
    }
 }
@@ -93,26 +93,33 @@ impl Layer for InMemoryLayer {
    // An in-memory layer doesn't really have a filename as it's not stored on disk,
    // but we construct a filename as if it was a delta layer
    fn filename(&self) -> PathBuf {
-        let inner = self.inner.read().unwrap();
+        let inner = self.inner.lock().unwrap();

        let end_lsn;
-        if let Some(drop_lsn) = inner.end_lsn {
+        let dropped;
+        if let Some(drop_lsn) = inner.drop_lsn {
            end_lsn = drop_lsn;
+            dropped = true;
        } else {
            end_lsn = Lsn(u64::MAX);
+            dropped = false;
        }

        let delta_filename = DeltaFileName {
            seg: self.seg,
            start_lsn: self.start_lsn,
            end_lsn,
-            dropped: inner.dropped,
+            dropped,
        }
        .to_string();

        PathBuf::from(format!("inmem-{}", delta_filename))
    }

+    fn path(&self) -> Option<PathBuf> {
+        None
+    }
+
    fn get_timeline_id(&self) -> ZTimelineId {
        self.timelineid
    }
@@ -126,18 +133,22 @@ impl Layer for InMemoryLayer {
    }

    fn get_end_lsn(&self) -> Lsn {
-        let inner = self.inner.read().unwrap();
+        if let Some(end_lsn) = self.end_lsn {
+            return Lsn(end_lsn.0 + 1);
+        }

-        if let Some(end_lsn) = inner.end_lsn {
-            end_lsn
+        let inner = self.inner.lock().unwrap();
+
+        if let Some(drop_lsn) = inner.drop_lsn {
+            drop_lsn
        } else {
            Lsn(u64::MAX)
        }
    }

    fn is_dropped(&self) -> bool {
-        let inner = self.inner.read().unwrap();
-        inner.dropped
+        let inner = self.inner.lock().unwrap();
+        inner.drop_lsn.is_some()
    }

    /// Look up given page in the cache.
@@ -151,22 +162,24 @@ impl Layer for InMemoryLayer {

        assert!(self.seg.blknum_in_seg(blknum));

-        {
-            let inner = self.inner.read().unwrap();
+        let predecessor: Option<Arc<dyn Layer>>;

-            // Scan the page versions backwards, starting from `lsn`.
-            let iter = inner
+        {
+            let inner = self.inner.lock().unwrap();
+
+            // Scan the BTreeMap backwards, starting from reconstruct_data.lsn.
+            let minkey = (blknum, Lsn(0));
+            let maxkey = (blknum, lsn);
+            let mut iter = inner
                .page_versions
-                .get_block_lsn_range(blknum, ..=lsn)
-                .iter()
-                .rev();
-            for (entry_lsn, entry) in iter {
+                .range((Included(&minkey), Included(&maxkey)));
+            while let Some(((_blknum, _entry_lsn), entry)) = iter.next_back() {
                if let Some(img) = &entry.page_image {
                    reconstruct_data.page_img = Some(img.clone());
                    need_image = false;
                    break;
                } else if let Some(rec) = &entry.record {
-                    reconstruct_data.records.push((*entry_lsn, rec.clone()));
+                    reconstruct_data.records.push(rec.clone());
                    if rec.will_init {
                        // This WAL record initializes the page, so no need to go further back
                        need_image = false;
@@ -177,14 +190,16 @@ impl Layer for InMemoryLayer {
                    bail!("no page image or WAL record for requested page");
                }
            }
+
+            predecessor = inner.predecessor.clone();
            // release lock on 'inner'
        }

        // If an older page image is needed to reconstruct the page, let the
-        // caller know
+        // caller know about the predecessor layer.
        if need_image {
-            if self.incremental {
-                Ok(PageReconstructResult::Continue(Lsn(self.start_lsn.0 - 1)))
+            if let Some(cont_layer) = predecessor {
+                Ok(PageReconstructResult::Continue(self.start_lsn, cont_layer))
            } else {
                Ok(PageReconstructResult::Missing(self.start_lsn))
            }
@@ -196,18 +211,14 @@ impl Layer for InMemoryLayer {
    /// Get size of the relation at given LSN
    fn get_seg_size(&self, lsn: Lsn) -> Result<u32> {
        assert!(lsn >= self.start_lsn);
-        ensure!(
-            self.seg.rel.is_blocky(),
-            "get_seg_size() called on a non-blocky rel"
-        );

-        let inner = self.inner.read().unwrap();
+        let inner = self.inner.lock().unwrap();
        Ok(inner.get_seg_size(lsn))
    }

    /// Does this segment exist at given LSN?
    fn get_seg_exists(&self, lsn: Lsn) -> Result<bool> {
-        let inner = self.inner.read().unwrap();
+        let inner = self.inner.lock().unwrap();

        // If the segment created after requested LSN,
        // it doesn't exist in the layer. But we shouldn't
@@ -215,8 +226,8 @@ impl Layer for InMemoryLayer {
        assert!(lsn >= self.start_lsn);

        // Is the requested LSN after the segment was dropped?
-        if let Some(end_lsn) = inner.end_lsn {
-            if lsn >= end_lsn {
+        if let Some(drop_lsn) = inner.drop_lsn {
+            if lsn >= drop_lsn {
                return Ok(false);
            }
        }
@@ -239,35 +250,36 @@ impl Layer for InMemoryLayer {
    }

    fn is_incremental(&self) -> bool {
-        self.incremental
+        let inner = self.inner.lock().unwrap();
+        inner.predecessor.is_some()
    }

    /// debugging function to print out the contents of the layer
    fn dump(&self) -> Result<()> {
-        let inner = self.inner.read().unwrap();
+        let inner = self.inner.lock().unwrap();

        let end_str = inner
-            .end_lsn
+            .drop_lsn
            .as_ref()
-            .map(Lsn::to_string)
+            .map(|drop_lsn| drop_lsn.to_string())
            .unwrap_or_default();

        println!(
-            "----- in-memory layer for tli {} seg {} {}-{} {} ----",
-            self.timelineid, self.seg, self.start_lsn, end_str, inner.dropped,
+            "----- in-memory layer for tli {} seg {} {}-{} ----",
+            self.timelineid, self.seg, self.start_lsn, end_str
        );

-        for (k, v) in inner.segsizes.as_slice() {
+        for (k, v) in inner.segsizes.iter() {
            println!("segsizes {}: {}", k, v);
        }

-        for (blknum, lsn, pv) in inner.page_versions.ordered_page_version_iter(None) {
+        for (k, v) in inner.page_versions.iter() {
            println!(
                "blk {} at {}: {}/{}\n",
-                blknum,
-                lsn,
-                pv.page_image.is_some(),
-                pv.record.is_some()
+                k.0,
+                k.1,
+                v.page_image.is_some(),
+                v.record.is_some()
            );
        }

@@ -275,19 +287,26 @@ impl Layer for InMemoryLayer {
    }
 }

-/// A result of an inmemory layer data being written to disk.
-pub struct LayersOnDisk {
-    pub delta_layers: Vec<DeltaLayer>,
-    pub image_layers: Vec<ImageLayer>,
-}
+/// Write failed because the layer is in process of being replaced.
+/// See [`LayeredTimeline::perform_write_op`] for how to handle this error.
+#[derive(Debug)]
+pub struct NonWriteableError;

-impl LayersOnDisk {
-    pub fn is_empty(&self) -> bool {
-        self.delta_layers.is_empty() && self.image_layers.is_empty()
-    }
+pub type WriteResult<T> = std::result::Result<T, NonWriteableError>;
+
+/// Helper struct to cleanup `InMemoryLayer::freeze` return signature.
+pub struct FreezeLayers {
+    /// Replacement layer for the layer which freeze was called on.
+    pub frozen: Arc<InMemoryLayer>,
+    /// New open layer containing leftover data.
+    pub open: Option<Arc<InMemoryLayer>>,
 }

 impl InMemoryLayer {
+    fn assert_not_frozen(&self) {
+        assert!(self.end_lsn.is_none());
+    }
+
    /// Return the oldest page version that's stored in this layer
    pub fn get_oldest_pending_lsn(&self) -> Lsn {
        self.oldest_pending_lsn
@@ -311,25 +330,20 @@ impl InMemoryLayer {
            start_lsn
        );

-        // The segment is initially empty, so initialize 'segsizes' with 0.
-        let mut segsizes = VecMap::default();
-        if seg.rel.is_blocky() {
-            segsizes.append(start_lsn, 0).unwrap();
-        }
-
        Ok(InMemoryLayer {
            conf,
            timelineid,
            tenantid,
            seg,
            start_lsn,
+            end_lsn: None,
            oldest_pending_lsn,
-            incremental: false,
-            inner: RwLock::new(InMemoryLayerInner {
-                end_lsn: None,
-                dropped: false,
-                page_versions: PageVersions::default(),
-                segsizes,
+            inner: Mutex::new(InMemoryLayerInner {
+                drop_lsn: None,
+                page_versions: BTreeMap::new(),
+                segsizes: BTreeMap::new(),
+                writeable: true,
+                predecessor: None,
            }),
        })
    }
@@ -337,10 +351,10 @@ impl InMemoryLayer {
    // Write operations

    /// Remember new page version, as a WAL record over previous version
-    pub fn put_wal_record(&self, lsn: Lsn, blknum: u32, rec: WALRecord) -> u32 {
+    pub fn put_wal_record(&self, blknum: u32, rec: WALRecord) -> WriteResult<u32> {
        self.put_page_version(
            blknum,
-            lsn,
+            rec.lsn,
            PageVersion {
                page_image: None,
                record: Some(rec),
@@ -349,7 +363,7 @@ impl InMemoryLayer {
    }

    /// Remember new page version, as a full page image
-    pub fn put_page_image(&self, blknum: u32, lsn: Lsn, img: Bytes) -> u32 {
+    pub fn put_page_image(&self, blknum: u32, lsn: Lsn, img: Bytes) -> WriteResult<u32> {
        self.put_page_version(
            blknum,
            lsn,
@@ -362,7 +376,8 @@ impl InMemoryLayer {

    /// Common subroutine of the public put_wal_record() and put_page_image() functions.
    /// Adds the page version to the in-memory tree
-    pub fn put_page_version(&self, blknum: u32, lsn: Lsn, pv: PageVersion) -> u32 {
+    pub fn put_page_version(&self, blknum: u32, lsn: Lsn, pv: PageVersion) -> WriteResult<u32> {
+        self.assert_not_frozen();
        assert!(self.seg.blknum_in_seg(blknum));

        trace!(
@@ -372,11 +387,11 @@ impl InMemoryLayer {
            self.timelineid,
            lsn
        );
-        let mut inner = self.inner.write().unwrap();
+        let mut inner = self.inner.lock().unwrap();

-        inner.assert_writeable();
+        inner.check_writeable()?;

-        let old = inner.page_versions.append_or_update_last(blknum, lsn, pv);
+        let old = inner.page_versions.insert((blknum, lsn), pv);

        if old.is_some() {
            // We already had an entry for this LSN. That's odd..
@@ -390,7 +405,7 @@ impl InMemoryLayer {
        if self.seg.rel.is_blocky() {
            let newsize = blknum - self.seg.segno * RELISH_SEG_SIZE + 1;

-            // use inner get_seg_size, since calling self.get_seg_size will try to acquire the lock,
+            // use inner get_seg_size, since calling self.get_seg_size will try to acquire self.inner.lock
            // which we've just acquired above
            let oldsize = inner.get_seg_size(lsn);
            if newsize > oldsize {
@@ -421,9 +436,7 @@ impl InMemoryLayer {
                        gapblknum,
                        blknum
                    );
-                    let old = inner
-                        .page_versions
-                        .append_or_update_last(gapblknum, lsn, zeropv);
+                    let old = inner.page_versions.insert((gapblknum, lsn), zeropv);
                    // We already had an entry for this LSN. That's odd..

                    if old.is_some() {
@@ -434,47 +447,48 @@ impl InMemoryLayer {
                    }
                }

-                inner.segsizes.append_or_update_last(lsn, newsize).unwrap();
-                return newsize - oldsize;
+                inner.segsizes.insert(lsn, newsize);
+                return Ok(newsize - oldsize);
            }
        }
-
-        0
+        Ok(0)
    }

    /// Remember that the relation was truncated at given LSN
-    pub fn put_truncation(&self, lsn: Lsn, segsize: u32) {
-        assert!(
-            self.seg.rel.is_blocky(),
-            "put_truncation() called on a non-blocky rel"
-        );
+    pub fn put_truncation(&self, lsn: Lsn, segsize: u32) -> WriteResult<()> {
+        self.assert_not_frozen();

-        let mut inner = self.inner.write().unwrap();
-        inner.assert_writeable();
+        let mut inner = self.inner.lock().unwrap();
+        inner.check_writeable()?;

        // check that this we truncate to a smaller size than segment was before the truncation
        let oldsize = inner.get_seg_size(lsn);
        assert!(segsize < oldsize);

-        let old = inner.segsizes.append_or_update_last(lsn, segsize).unwrap();
+        let old = inner.segsizes.insert(lsn, segsize);

        if old.is_some() {
            // We already had an entry for this LSN. That's odd..
            warn!("Inserting truncation, but had an entry for the LSN already");
        }
+
+        Ok(())
    }

    /// Remember that the segment was dropped at given LSN
-    pub fn drop_segment(&self, lsn: Lsn) {
-        let mut inner = self.inner.write().unwrap();
+    pub fn drop_segment(&self, lsn: Lsn) -> WriteResult<()> {
+        self.assert_not_frozen();

-        assert!(inner.end_lsn.is_none());
-        assert!(!inner.dropped);
-        inner.dropped = true;
-        assert!(self.start_lsn < lsn);
-        inner.end_lsn = Some(lsn);
+        let mut inner = self.inner.lock().unwrap();

-        trace!("dropped segment {} at {}", self.seg, lsn);
+        inner.check_writeable()?;
+
+        assert!(inner.drop_lsn.is_none());
+        inner.drop_lsn = Some(lsn);
+
+        info!("dropped segment {} at {}", self.seg, lsn);
+
+        Ok(())
    }

    ///
@@ -491,9 +505,6 @@ impl InMemoryLayer {
    ) -> Result<InMemoryLayer> {
        let seg = src.get_seg_tag();

-        assert!(oldest_pending_lsn.is_aligned());
-        assert!(oldest_pending_lsn >= start_lsn);
-
        trace!(
            "initializing new InMemoryLayer for writing {} on timeline {} at {}",
            seg,
@@ -501,11 +512,11 @@ impl InMemoryLayer {
            start_lsn,
        );

-        // Copy the segment size at the start LSN from the predecessor layer.
-        let mut segsizes = VecMap::default();
+        // For convenience, copy the segment size from the predecessor layer
+        let mut segsizes = BTreeMap::new();
        if seg.rel.is_blocky() {
            let size = src.get_seg_size(start_lsn)?;
-            segsizes.append(start_lsn, size).unwrap();
+            segsizes.insert(start_lsn, size);
        }

        Ok(InMemoryLayer {
@@ -514,43 +525,117 @@ impl InMemoryLayer {
            tenantid,
            seg,
            start_lsn,
+            end_lsn: None,
            oldest_pending_lsn,
-            incremental: true,
-            inner: RwLock::new(InMemoryLayerInner {
-                end_lsn: None,
-                dropped: false,
-                page_versions: PageVersions::default(),
+            inner: Mutex::new(InMemoryLayerInner {
+                drop_lsn: None,
+                page_versions: BTreeMap::new(),
                segsizes,
+                writeable: true,
+                predecessor: Some(src),
            }),
        })
    }

-    pub fn is_writeable(&self) -> bool {
-        let inner = self.inner.read().unwrap();
-        inner.end_lsn.is_none()
-    }
+    /// Splits `self` into two InMemoryLayers: `frozen` and `open`.
+    /// All data up to and including `cutoff_lsn` (or the drop LSN, if dropped)
+    /// is copied to `frozen`, while the remaining data is copied to `open`.
+    /// After completion, self is non-writeable, but not frozen.
+    pub fn freeze(&self, cutoff_lsn: Lsn) -> Result<FreezeLayers> {
+        info!(
+            "freezing in memory layer for {} on timeline {} at {}",
+            self.seg, self.timelineid, cutoff_lsn
+        );

-    /// Make the layer non-writeable. Only call once.
-    /// Records the end_lsn for non-dropped layers.
-    /// `end_lsn` is inclusive
-    pub fn freeze(&self, end_lsn: Lsn) {
-        let mut inner = self.inner.write().unwrap();
+        self.assert_not_frozen();

-        if inner.end_lsn.is_some() {
-            assert!(inner.dropped);
+        let mut inner = self.inner.lock().unwrap();
+        assert!(inner.writeable);
+        inner.writeable = false;
+
+        // Normally, use the cutoff LSN as the end of the frozen layer.
+        // But if the relation was dropped, we know that there are no
+        // more changes coming in for it, and in particular we know that
+        // there are no changes "in flight" for the LSN anymore, so we use
+        // the drop LSN instead. The drop-LSN could be ahead of the
+        // caller-specified LSN!
+        let dropped = inner.drop_lsn.is_some();
+        let end_lsn = if dropped {
+            inner.drop_lsn.unwrap()
        } else {
-            assert!(!inner.dropped);
-            assert!(self.start_lsn < end_lsn + 1);
-            inner.end_lsn = Some(Lsn(end_lsn.0 + 1));
+            cutoff_lsn
+        };

-            if let Some((lsn, _)) = inner.segsizes.as_slice().last() {
-                assert!(lsn <= &end_lsn, "{:?} {:?}", lsn, end_lsn);
+        // Divide all the page versions into old and new at the 'end_lsn' cutoff point.
+        let mut before_page_versions;
+        let mut before_segsizes;
+        let mut after_page_versions;
+        let mut after_segsizes;
+        if !dropped {
+            before_segsizes = BTreeMap::new();
+            after_segsizes = BTreeMap::new();
+            for (lsn, size) in inner.segsizes.iter() {
+                if *lsn > end_lsn {
+                    after_segsizes.insert(*lsn, *size);
+                } else {
+                    before_segsizes.insert(*lsn, *size);
+                }
            }

-            for (_blk, lsn, _pv) in inner.page_versions.ordered_page_version_iter(None) {
-                assert!(lsn <= end_lsn);
+            before_page_versions = BTreeMap::new();
+            after_page_versions = BTreeMap::new();
+            for ((blknum, lsn), pv) in inner.page_versions.iter() {
+                if *lsn > end_lsn {
+                    after_page_versions.insert((*blknum, *lsn), pv.clone());
+                } else {
+                    before_page_versions.insert((*blknum, *lsn), pv.clone());
+                }
            }
+        } else {
+            before_page_versions = inner.page_versions.clone();
+            before_segsizes = inner.segsizes.clone();
+            after_segsizes = BTreeMap::new();
+            after_page_versions = BTreeMap::new();
        }
+
+        let frozen = Arc::new(InMemoryLayer {
+            conf: self.conf,
+            tenantid: self.tenantid,
+            timelineid: self.timelineid,
+            seg: self.seg,
+            start_lsn: self.start_lsn,
+            end_lsn: Some(end_lsn),
+            oldest_pending_lsn: self.start_lsn,
+            inner: Mutex::new(InMemoryLayerInner {
+                drop_lsn: inner.drop_lsn,
+                page_versions: before_page_versions,
+                segsizes: before_segsizes,
+                writeable: false,
+                predecessor: inner.predecessor.clone(),
+            }),
+        });
+
+        let open = if !dropped && (!after_segsizes.is_empty() || !after_page_versions.is_empty()) {
+            let mut new_open = Self::create_successor_layer(
+                self.conf,
+                frozen.clone(),
+                self.timelineid,
+                self.tenantid,
+                end_lsn,
+                end_lsn,
+            )?;
+
+            let new_inner = new_open.inner.get_mut().unwrap();
+            new_inner.page_versions.append(&mut after_page_versions);
+            new_inner.segsizes.append(&mut after_segsizes);
+
+            Some(Arc::new(new_open))
+        } else {
+            None
+        };
+
+        // TODO could we avoid creating the `frozen` if it contains no data
+        Ok(FreezeLayers { frozen, open })
    }

    /// Write the this frozen in-memory layer to disk.
@@ -561,62 +646,40 @@ impl InMemoryLayer {
    /// WAL records between start and end LSN. (The delta layer is not needed
    /// when a new relish is created with a single LSN, so that the start and
    /// end LSN are the same.)
-    pub fn write_to_disk(&self, timeline: &LayeredTimeline) -> Result<LayersOnDisk> {
-        trace!(
-            "write_to_disk {} get_end_lsn is {}",
-            self.filename().display(),
-            self.get_end_lsn()
-        );
+    pub fn write_to_disk(&self, timeline: &LayeredTimeline) -> Result<Vec<Arc<dyn Layer>>> {
+        let end_lsn = self.end_lsn.expect("can only write frozen layers to disk");

-        // Grab the lock in read-mode. We hold it over the I/O, but because this
-        // layer is not writeable anymore, no one should be trying to acquire the
-        // write lock on it, so we shouldn't block anyone. There's one exception
-        // though: another thread might have grabbed a reference to this layer
-        // in `get_layer_for_write' just before the checkpointer called
-        // `freeze`, and then `write_to_disk` on it. When the thread gets the
-        // lock, it will see that it's not writeable anymore and retry, but it
-        // would have to wait until we release it. That race condition is very
-        // rare though, so we just accept the potential latency hit for now.
-        let inner = self.inner.read().unwrap();
-        let end_lsn_exclusive = inner.end_lsn.unwrap();
+        let inner = self.inner.lock().unwrap();

-        if inner.dropped {
-            let delta_layer = DeltaLayer::create(
-                self.conf,
-                self.timelineid,
-                self.tenantid,
-                self.seg,
-                self.start_lsn,
-                end_lsn_exclusive,
-                true,
-                inner.page_versions.ordered_page_version_iter(None),
-                inner.segsizes.clone(),
-            )?;
-            trace!(
-                "freeze: created delta layer for dropped segment {} {}-{}",
-                self.seg,
-                self.start_lsn,
-                end_lsn_exclusive
-            );
-            return Ok(LayersOnDisk {
-                delta_layers: vec![delta_layer],
-                image_layers: Vec::new(),
-            });
+        let drop_lsn = inner.drop_lsn;
+        let predecessor = inner.predecessor.clone();
+
+        let mut before_page_versions;
+        let mut before_segsizes;
+        if inner.drop_lsn.is_none() {
+            before_segsizes = BTreeMap::new();
+            for (lsn, size) in inner.segsizes.iter() {
+                if *lsn <= end_lsn {
+                    before_segsizes.insert(*lsn, *size);
+                }
+            }
+
+            before_page_versions = BTreeMap::new();
+            for ((blknum, lsn), pv) in inner.page_versions.iter() {
+                if *lsn < end_lsn {
+                    before_page_versions.insert((*blknum, *lsn), pv.clone());
+                }
+            }
+        } else {
+            before_page_versions = inner.page_versions.clone();
+            before_segsizes = inner.segsizes.clone();
        }

-        // Since `end_lsn` is inclusive, subtract 1.
-        // We want to make an ImageLayer for the last included LSN,
-        // so the DeltaLayer should exlcude that LSN.
-        let end_lsn_inclusive = Lsn(end_lsn_exclusive.0 - 1);
+        drop(inner);

-        let mut page_versions = inner
-            .page_versions
-            .ordered_page_version_iter(Some(end_lsn_inclusive));
+        let mut frozen_layers: Vec<Arc<dyn Layer>> = Vec::new();

-        let mut delta_layers = Vec::new();
-
-        if self.start_lsn != end_lsn_inclusive {
-            let (segsizes, _) = inner.segsizes.split_at(&end_lsn_exclusive);
+        if self.start_lsn != end_lsn {
            // Write the page versions before the cutoff to disk.
            let delta_layer = DeltaLayer::create(
                self.conf,
@@ -624,36 +687,35 @@ impl InMemoryLayer {
                self.tenantid,
                self.seg,
                self.start_lsn,
-                end_lsn_inclusive,
-                false,
-                page_versions,
-                segsizes,
+                end_lsn,
+                drop_lsn.is_some(),
+                predecessor,
+                before_page_versions,
+                before_segsizes,
            )?;
-            delta_layers.push(delta_layer);
+            frozen_layers.push(Arc::new(delta_layer));
            trace!(
                "freeze: created delta layer {} {}-{}",
                self.seg,
                self.start_lsn,
-                end_lsn_inclusive
+                end_lsn
            );
        } else {
-            assert!(page_versions.next().is_none());
+            assert!(before_page_versions.is_empty());
        }

-        drop(inner);
+        if drop_lsn.is_none() {
+            // Write a new base image layer at the cutoff point
+            let image_layer = ImageLayer::create_from_src(self.conf, timeline, self, end_lsn)?;
+            frozen_layers.push(Arc::new(image_layer));
+            trace!("freeze: created image layer {} at {}", self.seg, end_lsn);
+        }

-        // Write a new base image layer at the cutoff point
-        let image_layer =
-            ImageLayer::create_from_src(self.conf, timeline, self, end_lsn_inclusive)?;
-        trace!(
-            "freeze: created image layer {} at {}",
-            self.seg,
-            end_lsn_inclusive
-        );
+        Ok(frozen_layers)
+    }

-        Ok(LayersOnDisk {
-            delta_layers,
-            image_layers: vec![image_layer],
-        })
+    pub fn update_predecessor(&self, predecessor: Arc<dyn Layer>) -> Option<Arc<dyn Layer>> {
+        let mut inner = self.inner.lock().unwrap();
+        inner.predecessor.replace(predecessor)
    }
 }
--- a/pageserver/src/layered_repository/interval_tree.rs
+++ b/pageserver/src/layered_repository/interval_tree.rs
@@ -1,468 +0,0 @@
-///
-/// IntervalTree is data structure for holding intervals. It is generic
-/// to make unit testing possible, but the only real user of it is the layer map,
-///
-/// It's inspired by the "segment tree" or a "statistic tree" as described in
-/// https://en.wikipedia.org/wiki/Segment_tree. However, we use a B-tree to hold
-/// the points instead of a binary tree. This is called an "interval tree" instead
-/// of "segment tree" because the term "segment" is already using Zenith to mean
-/// something else. To add to the confusion, there is another data structure known
-/// as "interval tree" out there (see https://en.wikipedia.org/wiki/Interval_tree),
-/// for storing intervals, but this isn't that.
-///
-/// The basic idea is to have a B-tree of "interesting Points". At each Point,
-/// there is a list of intervals that contain the point. The Points are formed
-/// from the start bounds of each interval; there is a Point for each distinct
-/// start bound.
-///
-/// Operations:
-///
-/// To find intervals that contain a given point, you search the b-tree to find
-/// the nearest Point <= search key. Then you just return the list of intervals.
-///
-/// To insert an interval, find the Point with start key equal to the inserted item.
-/// If the Point doesn't exist yet, create it, by copying all the items from the
-/// previous Point that cover the new Point. Then walk right, inserting the new
-/// interval to all the Points that are contained by the new interval (including the
-/// newly created Point).
-///
-/// To remove an interval, you scan the tree for all the Points that are contained by
-/// the removed interval, and remove it from the list in each Point.
-///
-/// Requirements and assumptions:
-///
-/// - Can store overlapping items
-/// - But there are not many overlapping items
-/// - The interval bounds don't change after it is added to the tree
-/// - Intervals are uniquely identified by pointer equality. You must not be insert the
-///   same interval object twice, and `remove` uses pointer equality to remove the right
-///   interval. It is OK to have two intervals with the same bounds, however.
-///
-use std::collections::BTreeMap;
-use std::fmt::Debug;
-use std::ops::Range;
-use std::sync::Arc;
-
-pub struct IntervalTree<I: ?Sized>
-where
-    I: IntervalItem,
-{
-    points: BTreeMap<I::Key, Point<I>>,
-}
-
-struct Point<I: ?Sized> {
-    /// All intervals that contain this point, in no particular order.
-    ///
-    /// We assume that there aren't a lot of overlappingg intervals, so that this vector
-    /// never grows very large. If that assumption doesn't hold, we could keep this ordered
-    /// by the end bound, to speed up `search`. But as long as there are only a few elements,
-    /// a linear search is OK.
-    elements: Vec<Arc<I>>,
-}
-
-/// Abstraction for an interval that can be stored in the tree
-///
-/// The start bound is inclusive and the end bound is exclusive. End must be greater
-/// than start.
-pub trait IntervalItem {
-    type Key: Ord + Copy + Debug + Sized;
-
-    fn start_key(&self) -> Self::Key;
-    fn end_key(&self) -> Self::Key;
-
-    fn bounds(&self) -> Range<Self::Key> {
-        self.start_key()..self.end_key()
-    }
-}
-
-impl<I: ?Sized> IntervalTree<I>
-where
-    I: IntervalItem,
-{
-    /// Return an element that contains 'key', or precedes it.
-    ///
-    /// If there are multiple candidates, returns the one with the highest 'end' key.
-    pub fn search(&self, key: I::Key) -> Option<Arc<I>> {
-        // Find the greatest point that precedes or is equal to the search key. If there is
-        // none, returns None.
-        let (_, p) = self.points.range(..=key).next_back()?;
-
-        // Find the element with the highest end key at this point
-        let highest_item = p
-            .elements
-            .iter()
-            .reduce(|a, b| {
-                // starting with Rust 1.53, could use `std::cmp::min_by_key` here
-                if a.end_key() > b.end_key() {
-                    a
-                } else {
-                    b
-                }
-            })
-            .unwrap();
-        Some(Arc::clone(highest_item))
-    }
-
-    /// Iterate over all items with start bound >= 'key'
-    pub fn iter_newer(&self, key: I::Key) -> IntervalIter<I> {
-        IntervalIter {
-            point_iter: self.points.range(key..),
-            elem_iter: None,
-        }
-    }
-
-    /// Iterate over all items
-    pub fn iter(&self) -> IntervalIter<I> {
-        IntervalIter {
-            point_iter: self.points.range(..),
-            elem_iter: None,
-        }
-    }
-
-    pub fn insert(&mut self, item: Arc<I>) {
-        let start_key = item.start_key();
-        let end_key = item.end_key();
-        assert!(start_key < end_key);
-        let bounds = start_key..end_key;
-
-        // Find the starting point and walk forward from there
-        let mut found_start_point = false;
-        let iter = self.points.range_mut(bounds);
-        for (point_key, point) in iter {
-            if *point_key == start_key {
-                found_start_point = true;
-                // It is an error to insert the same item to the tree twice.
-                assert!(
-                    !point.elements.iter().any(|x| Arc::ptr_eq(x, &item)),
-                    "interval is already in the tree"
-                );
-            }
-            point.elements.push(Arc::clone(&item));
-        }
-        if !found_start_point {
-            // Create a new Point for the starting point
-
-            // Look at the previous point, and copy over elements that overlap with this
-            // new point
-            let mut new_elements: Vec<Arc<I>> = Vec::new();
-            if let Some((_, prev_point)) = self.points.range(..start_key).next_back() {
-                let overlapping_prev_elements = prev_point
-                    .elements
-                    .iter()
-                    .filter(|x| x.bounds().contains(&start_key))
-                    .cloned();
-
-                new_elements.extend(overlapping_prev_elements);
-            }
-            new_elements.push(item);
-
-            let new_point = Point {
-                elements: new_elements,
-            };
-            self.points.insert(start_key, new_point);
-        }
-    }
-
-    pub fn remove(&mut self, item: &Arc<I>) {
-        // range search points
-        let start_key = item.start_key();
-        let end_key = item.end_key();
-        let bounds = start_key..end_key;
-
-        let mut points_to_remove: Vec<I::Key> = Vec::new();
-        let mut found_start_point = false;
-        for (point_key, point) in self.points.range_mut(bounds) {
-            if *point_key == start_key {
-                found_start_point = true;
-            }
-            let len_before = point.elements.len();
-            point.elements.retain(|other| !Arc::ptr_eq(other, item));
-            let len_after = point.elements.len();
-            assert_eq!(len_after + 1, len_before);
-            if len_after == 0 {
-                points_to_remove.push(*point_key);
-            }
-        }
-        assert!(found_start_point);
-
-        for k in points_to_remove {
-            self.points.remove(&k).unwrap();
-        }
-    }
-}
-
-pub struct IntervalIter<'a, I: ?Sized>
-where
-    I: IntervalItem,
-{
-    point_iter: std::collections::btree_map::Range<'a, I::Key, Point<I>>,
-    elem_iter: Option<(I::Key, std::slice::Iter<'a, Arc<I>>)>,
-}
-
-impl<'a, I> Iterator for IntervalIter<'a, I>
-where
-    I: IntervalItem + ?Sized,
-{
-    type Item = Arc<I>;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        // Iterate over all elements in all the points in 'point_iter'. To avoid
-        // returning the same element twice, we only return each element at its
-        // starting point.
-        loop {
-            // Return next remaining element from the current point
-            if let Some((point_key, elem_iter)) = &mut self.elem_iter {
-                for elem in elem_iter {
-                    if elem.start_key() == *point_key {
-                        return Some(Arc::clone(elem));
-                    }
-                }
-            }
-            // No more elements at this point. Move to next point.
-            if let Some((point_key, point)) = self.point_iter.next() {
-                self.elem_iter = Some((*point_key, point.elements.iter()));
-                continue;
-            } else {
-                // No more points, all done
-                return None;
-            }
-        }
-    }
-}
-
-impl<I: ?Sized> Default for IntervalTree<I>
-where
-    I: IntervalItem,
-{
-    fn default() -> Self {
-        IntervalTree {
-            points: BTreeMap::new(),
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use std::fmt;
-
-    #[derive(Debug)]
-    struct MockItem {
-        start_key: u32,
-        end_key: u32,
-        val: String,
-    }
-    impl IntervalItem for MockItem {
-        type Key = u32;
-
-        fn start_key(&self) -> u32 {
-            self.start_key
-        }
-        fn end_key(&self) -> u32 {
-            self.end_key
-        }
-    }
-    impl MockItem {
-        fn new(start_key: u32, end_key: u32) -> Self {
-            MockItem {
-                start_key,
-                end_key,
-                val: format!("{}-{}", start_key, end_key),
-            }
-        }
-        fn new_str(start_key: u32, end_key: u32, val: &str) -> Self {
-            MockItem {
-                start_key,
-                end_key,
-                val: format!("{}-{}: {}", start_key, end_key, val),
-            }
-        }
-    }
-    impl fmt::Display for MockItem {
-        fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-            write!(f, "{}", self.val)
-        }
-    }
-    #[rustfmt::skip]
-    fn assert_search(
-        tree: &IntervalTree<MockItem>,
-        key: u32,
-        expected: &[&str],
-    ) -> Option<Arc<MockItem>> {
-        if let Some(v) = tree.search(key) {
-            let vstr = v.to_string();
-
-            assert!(!expected.is_empty(), "search with {} returned {}, expected None", key, v);
-            assert!(
-                expected.contains(&vstr.as_str()),
-                "search with {} returned {}, expected one of: {:?}",
-                key, v, expected,
-            );
-
-            Some(v)
-        } else {
-            assert!(
-                expected.is_empty(),
-                "search with {} returned None, expected one of {:?}",
-                key, expected
-            );
-            None
-        }
-    }
-
-    fn assert_contents(tree: &IntervalTree<MockItem>, expected: &[&str]) {
-        let mut contents: Vec<String> = tree.iter().map(|e| e.to_string()).collect();
-        contents.sort();
-        assert_eq!(contents, expected);
-    }
-
-    fn dump_tree(tree: &IntervalTree<MockItem>) {
-        for (point_key, point) in tree.points.iter() {
-            print!("{}:", point_key);
-            for e in point.elements.iter() {
-                print!(" {}", e);
-            }
-            println!();
-        }
-    }
-
-    #[test]
-    fn test_interval_tree_simple() {
-        let mut tree: IntervalTree<MockItem> = IntervalTree::default();
-
-        // Simple, non-overlapping ranges.
-        tree.insert(Arc::new(MockItem::new(10, 11)));
-        tree.insert(Arc::new(MockItem::new(11, 12)));
-        tree.insert(Arc::new(MockItem::new(12, 13)));
-        tree.insert(Arc::new(MockItem::new(18, 19)));
-        tree.insert(Arc::new(MockItem::new(17, 18)));
-        tree.insert(Arc::new(MockItem::new(15, 16)));
-
-        assert_search(&tree, 9, &[]);
-        assert_search(&tree, 10, &["10-11"]);
-        assert_search(&tree, 11, &["11-12"]);
-        assert_search(&tree, 12, &["12-13"]);
-        assert_search(&tree, 13, &["12-13"]);
-        assert_search(&tree, 14, &["12-13"]);
-        assert_search(&tree, 15, &["15-16"]);
-        assert_search(&tree, 16, &["15-16"]);
-        assert_search(&tree, 17, &["17-18"]);
-        assert_search(&tree, 18, &["18-19"]);
-        assert_search(&tree, 19, &["18-19"]);
-        assert_search(&tree, 20, &["18-19"]);
-
-        // remove a few entries and search around them again
-        tree.remove(&assert_search(&tree, 10, &["10-11"]).unwrap()); // first entry
-        tree.remove(&assert_search(&tree, 12, &["12-13"]).unwrap()); // entry in the middle
-        tree.remove(&assert_search(&tree, 18, &["18-19"]).unwrap()); // last entry
-        assert_search(&tree, 9, &[]);
-        assert_search(&tree, 10, &[]);
-        assert_search(&tree, 11, &["11-12"]);
-        assert_search(&tree, 12, &["11-12"]);
-        assert_search(&tree, 14, &["11-12"]);
-        assert_search(&tree, 15, &["15-16"]);
-        assert_search(&tree, 17, &["17-18"]);
-        assert_search(&tree, 18, &["17-18"]);
-    }
-
-    #[test]
-    fn test_interval_tree_overlap() {
-        let mut tree: IntervalTree<MockItem> = IntervalTree::default();
-
-        // Overlapping items
-        tree.insert(Arc::new(MockItem::new(22, 24)));
-        tree.insert(Arc::new(MockItem::new(23, 25)));
-        let x24_26 = Arc::new(MockItem::new(24, 26));
-        tree.insert(Arc::clone(&x24_26));
-        let x26_28 = Arc::new(MockItem::new(26, 28));
-        tree.insert(Arc::clone(&x26_28));
-        tree.insert(Arc::new(MockItem::new(25, 27)));
-
-        assert_search(&tree, 22, &["22-24"]);
-        assert_search(&tree, 23, &["22-24", "23-25"]);
-        assert_search(&tree, 24, &["23-25", "24-26"]);
-        assert_search(&tree, 25, &["24-26", "25-27"]);
-        assert_search(&tree, 26, &["25-27", "26-28"]);
-        assert_search(&tree, 27, &["26-28"]);
-        assert_search(&tree, 28, &["26-28"]);
-        assert_search(&tree, 29, &["26-28"]);
-
-        tree.remove(&x24_26);
-        tree.remove(&x26_28);
-        assert_search(&tree, 23, &["22-24", "23-25"]);
-        assert_search(&tree, 24, &["23-25"]);
-        assert_search(&tree, 25, &["25-27"]);
-        assert_search(&tree, 26, &["25-27"]);
-        assert_search(&tree, 27, &["25-27"]);
-        assert_search(&tree, 28, &["25-27"]);
-        assert_search(&tree, 29, &["25-27"]);
-    }
-
-    #[test]
-    fn test_interval_tree_nested() {
-        let mut tree: IntervalTree<MockItem> = IntervalTree::default();
-
-        // Items containing other items
-        tree.insert(Arc::new(MockItem::new(31, 39)));
-        tree.insert(Arc::new(MockItem::new(32, 34)));
-        tree.insert(Arc::new(MockItem::new(33, 35)));
-        tree.insert(Arc::new(MockItem::new(30, 40)));
-
-        assert_search(&tree, 30, &["30-40"]);
-        assert_search(&tree, 31, &["30-40", "31-39"]);
-        assert_search(&tree, 32, &["30-40", "32-34", "31-39"]);
-        assert_search(&tree, 33, &["30-40", "32-34", "33-35", "31-39"]);
-        assert_search(&tree, 34, &["30-40", "33-35", "31-39"]);
-        assert_search(&tree, 35, &["30-40", "31-39"]);
-        assert_search(&tree, 36, &["30-40", "31-39"]);
-        assert_search(&tree, 37, &["30-40", "31-39"]);
-        assert_search(&tree, 38, &["30-40", "31-39"]);
-        assert_search(&tree, 39, &["30-40"]);
-        assert_search(&tree, 40, &["30-40"]);
-        assert_search(&tree, 41, &["30-40"]);
-    }
-
-    #[test]
-    fn test_interval_tree_duplicates() {
-        let mut tree: IntervalTree<MockItem> = IntervalTree::default();
-
-        // Duplicate keys
-        let item_a = Arc::new(MockItem::new_str(55, 56, "a"));
-        tree.insert(Arc::clone(&item_a));
-        let item_b = Arc::new(MockItem::new_str(55, 56, "b"));
-        tree.insert(Arc::clone(&item_b));
-        let item_c = Arc::new(MockItem::new_str(55, 56, "c"));
-        tree.insert(Arc::clone(&item_c));
-        let item_d = Arc::new(MockItem::new_str(54, 56, "d"));
-        tree.insert(Arc::clone(&item_d));
-        let item_e = Arc::new(MockItem::new_str(55, 57, "e"));
-        tree.insert(Arc::clone(&item_e));
-
-        dump_tree(&tree);
-
-        assert_search(
-            &tree,
-            55,
-            &["55-56: a", "55-56: b", "55-56: c", "54-56: d", "55-57: e"],
-        );
-        tree.remove(&item_b);
-        dump_tree(&tree);
-
-        assert_contents(&tree, &["54-56: d", "55-56: a", "55-56: c", "55-57: e"]);
-
-        tree.remove(&item_d);
-        dump_tree(&tree);
-        assert_contents(&tree, &["55-56: a", "55-56: c", "55-57: e"]);
-    }
-
-    #[test]
-    #[should_panic]
-    fn test_interval_tree_insert_twice() {
-        let mut tree: IntervalTree<MockItem> = IntervalTree::default();
-
-        // Inserting the same item twice is not cool
-        let item = Arc::new(MockItem::new(1, 2));
-        tree.insert(Arc::clone(&item));
-        tree.insert(Arc::clone(&item)); // fails assertion
-    }
-}
--- a/pageserver/src/layered_repository/layer_map.rs
+++ b/pageserver/src/layered_repository/layer_map.rs
@@ -9,14 +9,13 @@
 //! new image and delta layers and corresponding files are written to disk.
 //!

-use crate::layered_repository::interval_tree::{IntervalItem, IntervalIter, IntervalTree};
 use crate::layered_repository::storage_layer::{Layer, SegmentTag};
 use crate::layered_repository::InMemoryLayer;
 use crate::relish::*;
 use anyhow::Result;
 use lazy_static::lazy_static;
 use std::cmp::Ordering;
-use std::collections::{BinaryHeap, HashMap};
+use std::collections::{BTreeMap, BinaryHeap, HashMap};
 use std::sync::Arc;
 use zenith_metrics::{register_int_gauge, IntGauge};
 use zenith_utils::lsn::Lsn;
@@ -77,14 +76,7 @@ impl LayerMap {
    pub fn insert_open(&mut self, layer: Arc<InMemoryLayer>) {
        let segentry = self.segs.entry(layer.get_seg_tag()).or_default();

-        segentry.update_open(Arc::clone(&layer));
-
-        let oldest_pending_lsn = layer.get_oldest_pending_lsn();
-
-        // After a crash and restart, 'oldest_pending_lsn' of the oldest in-memory
-        // layer becomes the WAL streaming starting point, so it better not point
-        // in the middle of a WAL record.
-        assert!(oldest_pending_lsn.is_aligned());
+        segentry.insert_open(Arc::clone(&layer));

        // Also add it to the binary heap
        let open_layer_entry = OpenLayerEntry {
@@ -105,14 +97,11 @@ impl LayerMap {

        // Also remove it from the SegEntry of this segment
        let mut segentry = self.segs.get_mut(&segtag).unwrap();
-        if Arc::ptr_eq(segentry.open.as_ref().unwrap(), &oldest_entry.layer) {
-            segentry.open = None;
-        } else {
-            // We could have already updated segentry.open for
-            // dropped (non-writeable) layer. This is fine.
-            assert!(!oldest_entry.layer.is_writeable());
-            assert!(oldest_entry.layer.is_dropped());
-        }
+        assert!(Arc::ptr_eq(
+            segentry.open.as_ref().unwrap(),
+            &oldest_entry.layer
+        ));
+        segentry.open = None;

        NUM_INMEMORY_LAYERS.dec();
    }
@@ -132,11 +121,12 @@ impl LayerMap {
    ///
    /// This should be called when the corresponding file on disk has been deleted.
    ///
-    pub fn remove_historic(&mut self, layer: Arc<dyn Layer>) {
+    pub fn remove_historic(&mut self, layer: &dyn Layer) {
        let tag = layer.get_seg_tag();
+        let start_lsn = layer.get_start_lsn();

        if let Some(segentry) = self.segs.get_mut(&tag) {
-            segentry.historic.remove(&layer);
+            segentry.historic.remove(&start_lsn);
        }
        NUM_ONDISK_LAYERS.dec();
    }
@@ -154,7 +144,7 @@ impl LayerMap {
                        if (request_rel.spcnode == 0 || reltag.spcnode == request_rel.spcnode)
                            && (request_rel.dbnode == 0 || reltag.dbnode == request_rel.dbnode)
                        {
-                            if let Some(exists) = segentry.exists_at_lsn(lsn)? {
+                            if let Some(exists) = segentry.exists_at_lsn(lsn) {
                                rels.insert(seg.rel, exists);
                            }
                        }
@@ -162,7 +152,7 @@ impl LayerMap {
                }
                _ => {
                    if tag == None {
-                        if let Some(exists) = segentry.exists_at_lsn(lsn)? {
+                        if let Some(exists) = segentry.exists_at_lsn(lsn) {
                            rels.insert(seg.rel, exists);
                        }
                    }
@@ -184,20 +174,6 @@ impl LayerMap {
        }
    }

-    /// Is there any layer for given segment that is alive at the lsn?
-    ///
-    /// This is a public wrapper for SegEntry fucntion,
-    /// used for garbage collection, to determine if some alive layer
-    /// exists at the lsn. If so, we shouldn't delete a newer dropped layer
-    /// to avoid incorrectly making it visible.
-    pub fn layer_exists_at_lsn(&self, seg: SegmentTag, lsn: Lsn) -> Result<bool> {
-        Ok(if let Some(segentry) = self.segs.get(&seg) {
-            segentry.exists_at_lsn(lsn)?.unwrap_or(false)
-        } else {
-            false
-        })
-    }
-
    /// Return the oldest in-memory layer, along with its generation number.
    pub fn peek_oldest_open(&self) -> Option<(Arc<InMemoryLayer>, u64)> {
        self.open_layers
@@ -215,7 +191,7 @@ impl LayerMap {

    pub fn iter_historic_layers(&self) -> HistoricLayerIter {
        HistoricLayerIter {
-            seg_iter: self.segs.iter(),
+            segiter: self.segs.iter(),
            iter: None,
        }
    }
@@ -229,7 +205,7 @@ impl LayerMap {
                open.dump()?;
            }

-            for layer in segentry.historic.iter() {
+            for (_, layer) in segentry.historic.iter() {
                layer.dump()?;
            }
        }
@@ -238,40 +214,34 @@ impl LayerMap {
    }
 }

-impl IntervalItem for dyn Layer {
-    type Key = Lsn;
-
-    fn start_key(&self) -> Lsn {
-        self.get_start_lsn()
-    }
-    fn end_key(&self) -> Lsn {
-        self.get_end_lsn()
-    }
-}
-
 ///
 /// Per-segment entry in the LayerMap::segs hash map. Holds all the layers
 /// associated with the segment.
 ///
 /// The last layer that is open for writes is always an InMemoryLayer,
 /// and is kept in a separate field, because there can be only one for
-/// each segment. The older layers, stored on disk, are kept in an
-/// IntervalTree.
+/// each segment. The older layers, stored on disk, are kept in a
+/// BTreeMap keyed by the layer's start LSN.
 #[derive(Default)]
 struct SegEntry {
-    open: Option<Arc<InMemoryLayer>>,
-    historic: IntervalTree<dyn Layer>,
+    pub open: Option<Arc<InMemoryLayer>>,
+    pub historic: BTreeMap<Lsn, Arc<dyn Layer>>,
 }

 impl SegEntry {
    /// Does the segment exist at given LSN?
    /// Return None if object is not found in this SegEntry.
-    fn exists_at_lsn(&self, lsn: Lsn) -> Result<Option<bool>> {
-        if let Some(layer) = self.get(lsn) {
-            Ok(Some(layer.get_seg_exists(lsn)?))
-        } else {
-            Ok(None)
+    fn exists_at_lsn(&self, lsn: Lsn) -> Option<bool> {
+        if let Some(layer) = &self.open {
+            if layer.get_start_lsn() <= lsn && lsn <= layer.get_end_lsn() {
+                let exists = layer.get_seg_exists(lsn).ok()?;
+                return Some(exists);
+            }
+        } else if let Some((_, layer)) = self.historic.range(..=lsn).next_back() {
+            let exists = layer.get_seg_exists(lsn).ok()?;
+            return Some(exists);
        }
+        None
    }

    pub fn get(&self, lsn: Lsn) -> Option<Arc<dyn Layer>> {
@@ -282,30 +252,40 @@ impl SegEntry {
            }
        }

-        self.historic.search(lsn)
+        if let Some((_start_lsn, layer)) = self.historic.range(..=lsn).next_back() {
+            Some(Arc::clone(layer))
+        } else {
+            None
+        }
    }

    pub fn newer_image_layer_exists(&self, lsn: Lsn) -> bool {
        // We only check on-disk layers, because
        // in-memory layers are not durable

-        self.historic
-            .iter_newer(lsn)
-            .any(|layer| !layer.is_incremental())
+        for (_newer_lsn, layer) in self.historic.range(lsn..) {
+            // Ignore incremental layers.
+            if layer.is_incremental() {
+                continue;
+            }
+            if layer.get_end_lsn() > lsn {
+                return true;
+            } else {
+                continue;
+            }
+        }
+        false
    }

-    // Set new open layer for a SegEntry.
-    // It's ok to rewrite previous open layer,
-    // but only if it is not writeable anymore.
-    pub fn update_open(&mut self, layer: Arc<InMemoryLayer>) {
-        if let Some(prev_open) = &self.open {
-            assert!(!prev_open.is_writeable());
-        }
+    pub fn insert_open(&mut self, layer: Arc<InMemoryLayer>) {
+        assert!(self.open.is_none());
        self.open = Some(layer);
    }

    pub fn insert_historic(&mut self, layer: Arc<dyn Layer>) {
-        self.historic.insert(layer);
+        let start_lsn = layer.get_start_lsn();
+
+        self.historic.insert(start_lsn, layer);
    }
 }

@@ -344,8 +324,8 @@ impl Eq for OpenLayerEntry {}

 /// Iterator returned by LayerMap::iter_historic_layers()
 pub struct HistoricLayerIter<'a> {
-    seg_iter: std::collections::hash_map::Iter<'a, SegmentTag, SegEntry>,
-    iter: Option<IntervalIter<'a, dyn Layer>>,
+    segiter: std::collections::hash_map::Iter<'a, SegmentTag, SegEntry>,
+    iter: Option<std::collections::btree_map::Iter<'a, Lsn, Arc<dyn Layer>>>,
 }

 impl<'a> Iterator for HistoricLayerIter<'a> {
@@ -355,11 +335,11 @@ impl<'a> Iterator for HistoricLayerIter<'a> {
        loop {
            if let Some(x) = &mut self.iter {
                if let Some(x) = x.next() {
-                    return Some(Arc::clone(&x));
+                    return Some(Arc::clone(&*x.1));
                }
            }
-            if let Some((_tag, segentry)) = self.seg_iter.next() {
-                self.iter = Some(segentry.historic.iter());
+            if let Some(seg) = self.segiter.next() {
+                self.iter = Some(seg.1.historic.iter());
                continue;
            } else {
                return None;
@@ -414,14 +394,14 @@ mod tests {
        let mut layers = LayerMap::default();

        let gen1 = layers.increment_generation();
-        layers.insert_open(dummy_inmem_layer(conf, 0, Lsn(0x100), Lsn(0x100)));
-        layers.insert_open(dummy_inmem_layer(conf, 1, Lsn(0x100), Lsn(0x200)));
-        layers.insert_open(dummy_inmem_layer(conf, 2, Lsn(0x100), Lsn(0x120)));
-        layers.insert_open(dummy_inmem_layer(conf, 3, Lsn(0x100), Lsn(0x110)));
+        layers.insert_open(dummy_inmem_layer(conf, 0, Lsn(100), Lsn(100)));
+        layers.insert_open(dummy_inmem_layer(conf, 1, Lsn(100), Lsn(200)));
+        layers.insert_open(dummy_inmem_layer(conf, 2, Lsn(100), Lsn(120)));
+        layers.insert_open(dummy_inmem_layer(conf, 3, Lsn(100), Lsn(110)));

        let gen2 = layers.increment_generation();
-        layers.insert_open(dummy_inmem_layer(conf, 4, Lsn(0x100), Lsn(0x110)));
-        layers.insert_open(dummy_inmem_layer(conf, 5, Lsn(0x100), Lsn(0x100)));
+        layers.insert_open(dummy_inmem_layer(conf, 4, Lsn(100), Lsn(110)));
+        layers.insert_open(dummy_inmem_layer(conf, 5, Lsn(100), Lsn(100)));

        // A helper function (closure) to pop the next oldest open entry from the layer map,
        // and assert that it is what we'd expect
@@ -432,12 +412,12 @@ mod tests {
            layers.pop_oldest_open();
        };

-        assert_pop_layer(0, gen1); // 0x100
-        assert_pop_layer(5, gen2); // 0x100
-        assert_pop_layer(3, gen1); // 0x110
-        assert_pop_layer(4, gen2); // 0x110
-        assert_pop_layer(2, gen1); // 0x120
-        assert_pop_layer(1, gen1); // 0x200
+        assert_pop_layer(0, gen1); // 100
+        assert_pop_layer(5, gen2); // 100
+        assert_pop_layer(3, gen1); // 110
+        assert_pop_layer(4, gen2); // 110
+        assert_pop_layer(2, gen1); // 120
+        assert_pop_layer(1, gen1); // 200

        Ok(())
    }
--- a/pageserver/src/layered_repository/page_versions.rs
+++ b/pageserver/src/layered_repository/page_versions.rs
@@ -1,150 +0,0 @@
-use std::{collections::HashMap, ops::RangeBounds, slice};
-
-use zenith_utils::{lsn::Lsn, vec_map::VecMap};
-
-use super::storage_layer::PageVersion;
-
-const EMPTY_SLICE: &[(Lsn, PageVersion)] = &[];
-
-#[derive(Debug, Default)]
-pub struct PageVersions(HashMap<u32, VecMap<Lsn, PageVersion>>);
-
-impl PageVersions {
-    pub fn append_or_update_last(
-        &mut self,
-        blknum: u32,
-        lsn: Lsn,
-        page_version: PageVersion,
-    ) -> Option<PageVersion> {
-        let map = self.0.entry(blknum).or_insert_with(VecMap::default);
-        map.append_or_update_last(lsn, page_version).unwrap()
-    }
-
-    /// Get all [`PageVersion`]s in a block
-    pub fn get_block_slice(&self, blknum: u32) -> &[(Lsn, PageVersion)] {
-        self.0
-            .get(&blknum)
-            .map(VecMap::as_slice)
-            .unwrap_or(EMPTY_SLICE)
-    }
-
-    /// Get a range of [`PageVersions`] in a block
-    pub fn get_block_lsn_range<R: RangeBounds<Lsn>>(
-        &self,
-        blknum: u32,
-        range: R,
-    ) -> &[(Lsn, PageVersion)] {
-        self.0
-            .get(&blknum)
-            .map(|vec_map| vec_map.slice_range(range))
-            .unwrap_or(EMPTY_SLICE)
-    }
-
-    /// Iterate through [`PageVersion`]s in (block, lsn) order.
-    /// If a [`cutoff_lsn`] is set, only show versions with `lsn < cutoff_lsn`
-    pub fn ordered_page_version_iter(&self, cutoff_lsn: Option<Lsn>) -> OrderedPageVersionIter<'_> {
-        let mut ordered_blocks: Vec<u32> = self.0.keys().cloned().collect();
-        ordered_blocks.sort_unstable();
-
-        let slice = ordered_blocks
-            .first()
-            .map(|&blknum| self.get_block_slice(blknum))
-            .unwrap_or(EMPTY_SLICE);
-
-        OrderedPageVersionIter {
-            page_versions: self,
-            ordered_blocks,
-            cur_block_idx: 0,
-            cutoff_lsn,
-            cur_slice_iter: slice.iter(),
-        }
-    }
-}
-
-pub struct OrderedPageVersionIter<'a> {
-    page_versions: &'a PageVersions,
-
-    ordered_blocks: Vec<u32>,
-    cur_block_idx: usize,
-
-    cutoff_lsn: Option<Lsn>,
-
-    cur_slice_iter: slice::Iter<'a, (Lsn, PageVersion)>,
-}
-
-impl OrderedPageVersionIter<'_> {
-    fn is_lsn_before_cutoff(&self, lsn: &Lsn) -> bool {
-        if let Some(cutoff_lsn) = self.cutoff_lsn.as_ref() {
-            lsn < cutoff_lsn
-        } else {
-            true
-        }
-    }
-}
-
-impl<'a> Iterator for OrderedPageVersionIter<'a> {
-    type Item = (u32, Lsn, &'a PageVersion);
-
-    fn next(&mut self) -> Option<Self::Item> {
-        loop {
-            if let Some((lsn, page_version)) = self.cur_slice_iter.next() {
-                if self.is_lsn_before_cutoff(lsn) {
-                    let blknum = self.ordered_blocks[self.cur_block_idx];
-                    return Some((blknum, *lsn, page_version));
-                }
-            }
-
-            let next_block_idx = self.cur_block_idx + 1;
-            let blknum: u32 = *self.ordered_blocks.get(next_block_idx)?;
-            self.cur_block_idx = next_block_idx;
-            self.cur_slice_iter = self.page_versions.get_block_slice(blknum).iter();
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    const EMPTY_PAGE_VERSION: PageVersion = PageVersion {
-        page_image: None,
-        record: None,
-    };
-
-    #[test]
-    fn test_ordered_iter() {
-        let mut page_versions = PageVersions::default();
-        const BLOCKS: u32 = 1000;
-        const LSNS: u64 = 50;
-
-        for blknum in 0..BLOCKS {
-            for lsn in 0..LSNS {
-                let old = page_versions.append_or_update_last(blknum, Lsn(lsn), EMPTY_PAGE_VERSION);
-                assert!(old.is_none());
-            }
-        }
-
-        let mut iter = page_versions.ordered_page_version_iter(None);
-        for blknum in 0..BLOCKS {
-            for lsn in 0..LSNS {
-                let (actual_blknum, actual_lsn, _pv) = iter.next().unwrap();
-                assert_eq!(actual_blknum, blknum);
-                assert_eq!(Lsn(lsn), actual_lsn);
-            }
-        }
-        assert!(iter.next().is_none());
-        assert!(iter.next().is_none()); // should be robust against excessive next() calls
-
-        const CUTOFF_LSN: Lsn = Lsn(30);
-        let mut iter = page_versions.ordered_page_version_iter(Some(CUTOFF_LSN));
-        for blknum in 0..BLOCKS {
-            for lsn in 0..CUTOFF_LSN.0 {
-                let (actual_blknum, actual_lsn, _pv) = iter.next().unwrap();
-                assert_eq!(actual_blknum, blknum);
-                assert_eq!(Lsn(lsn), actual_lsn);
-            }
-        }
-        assert!(iter.next().is_none());
-        assert!(iter.next().is_none()); // should be robust against excessive next() calls
-    }
-}
--- a/pageserver/src/layered_repository/storage_layer.rs
+++ b/pageserver/src/layered_repository/storage_layer.rs
@@ -10,6 +10,7 @@ use bytes::Bytes;
 use serde::{Deserialize, Serialize};
 use std::fmt;
 use std::path::PathBuf;
+use std::sync::Arc;

 use zenith_utils::lsn::Lsn;

@@ -78,7 +79,7 @@ pub struct PageVersion {
 /// 'records' contains the records to apply over the base image.
 ///
 pub struct PageReconstructData {
-    pub records: Vec<(Lsn, WALRecord)>,
+    pub records: Vec<WALRecord>,
    pub page_img: Option<Bytes>,
 }

@@ -86,9 +87,9 @@ pub struct PageReconstructData {
 pub enum PageReconstructResult {
    /// Got all the data needed to reconstruct the requested page
    Complete,
-    /// This layer didn't contain all the required data, the caller should look up
-    /// the predecessor layer at the returned LSN and collect more data from there.
-    Continue(Lsn),
+    /// This layer didn't contain all the required data, the caller should collect
+    /// more data from the returned predecessor layer at the returned LSN.
+    Continue(Lsn, Arc<dyn Layer>),
    /// This layer didn't contain data needed to reconstruct the page version at
    /// the returned LSN. This is usually considered an error, but might be OK
    /// in some circumstances.
@@ -110,19 +111,24 @@ pub trait Layer: Send + Sync {
    /// Identify the relish segment
    fn get_seg_tag(&self) -> SegmentTag;

-    /// Inclusive start bound of the LSN range that this layer holds
+    /// Inclusive start bound of the LSN range that this layer hold
    fn get_start_lsn(&self) -> Lsn;

-    /// Exclusive end bound of the LSN range that this layer holds.
+    /// 'end_lsn' meaning depends on the layer kind:
+    /// - in-memory layer is either unbounded (end_lsn = MAX_LSN) or dropped (end_lsn = drop_lsn)
+    /// - image layer represents snapshot at one LSN, so end_lsn = lsn
+    /// - delta layer has end_lsn
    ///
-    /// - For an open in-memory layer, this is MAX_LSN.
-    /// - For a frozen in-memory layer or a delta layer, this is a valid end bound.
-    /// - An image layer represents snapshot at one LSN, so end_lsn is always the snapshot LSN + 1
+    /// TODO Is end_lsn always exclusive for all layer kinds?
    fn get_end_lsn(&self) -> Lsn;

    /// Is the segment represented by this layer dropped by PostgreSQL?
    fn is_dropped(&self) -> bool;

+    /// Gets the physical location of the layer on disk.
+    /// Some layers, such as in-memory, might not have the location.
+    fn path(&self) -> Option<PathBuf>;
+
    /// Filename used to store this layer on disk. (Even in-memory layers
    /// implement this, to print a handy unique identifier for the layer for
    /// log messages, even though they're never not on disk.)
@@ -140,8 +146,8 @@ pub trait Layer: Send + Sync {
    ///
    /// See PageReconstructResult for possible return values. The collected data
    /// is appended to reconstruct_data; the caller should pass an empty struct
-    /// on first call. If this returns PageReconstructResult::Continue, look up
-    /// the predecessor layer and call again with the same 'reconstruct_data'
+    /// on first call. If this returns PageReconstructResult::Continue, call
+    /// again on the returned predecessor layer with the same 'reconstruct_data'
    /// to collect more data.
    fn get_page_reconstruct_data(
        &self,
--- a/pageserver/src/lib.rs
+++ b/pageserver/src/lib.rs
@@ -13,7 +13,7 @@ pub mod http;
 pub mod layered_repository;
 pub mod page_service;
 pub mod relish;
-pub mod relish_storage;
+mod relish_storage;
 pub mod repository;
 pub mod restore_local_repo;
 pub mod tenant_mgr;
@@ -22,13 +22,12 @@ pub mod walreceiver;
 pub mod walredo;

 pub mod defaults {
-    use const_format::formatcp;
    use std::time::Duration;

    pub const DEFAULT_PG_LISTEN_PORT: u16 = 64000;
-    pub const DEFAULT_PG_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_PG_LISTEN_PORT}");
+    pub const DEFAULT_PG_LISTEN_ADDR: &str = "127.0.0.1:64000"; // can't format! const yet...
    pub const DEFAULT_HTTP_LISTEN_PORT: u16 = 9898;
-    pub const DEFAULT_HTTP_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_HTTP_LISTEN_PORT}");
+    pub const DEFAULT_HTTP_LISTEN_ADDR: &str = "127.0.0.1:9898";

    // FIXME: This current value is very low. I would imagine something like 1 GB or 10 GB
    // would be more appropriate. But a low value forces the code to be exercised more,
@@ -40,7 +39,6 @@ pub mod defaults {
    pub const DEFAULT_GC_PERIOD: Duration = Duration::from_secs(100);

    pub const DEFAULT_SUPERUSER: &str = "zenith_admin";
-    pub const DEFAULT_RELISH_STORAGE_MAX_CONCURRENT_SYNC_LIMITS: usize = 100;
 }

 lazy_static! {
@@ -126,6 +124,10 @@ impl PageServerConf {
        self.timeline_path(timelineid, tenantid).join("ancestor")
    }

+    fn wal_dir_path(&self, timelineid: &ZTimelineId, tenantid: &ZTenantId) -> PathBuf {
+        self.timeline_path(timelineid, tenantid).join("wal")
+    }
+
    //
    // Postgres distribution paths
    //
@@ -151,8 +153,8 @@ impl PageServerConf {
            checkpoint_period: Duration::from_secs(10),
            gc_horizon: defaults::DEFAULT_GC_HORIZON,
            gc_period: Duration::from_secs(10),
-            listen_pg_addr: defaults::DEFAULT_PG_LISTEN_ADDR.to_string(),
-            listen_http_addr: defaults::DEFAULT_HTTP_LISTEN_ADDR.to_string(),
+            listen_pg_addr: "127.0.0.1:5430".to_string(),
+            listen_http_addr: "127.0.0.1:9898".to_string(),
            superuser: "zenith_admin".to_string(),
            workdir: repo_dir,
            pg_distrib_dir: "".into(),
@@ -165,37 +167,18 @@ impl PageServerConf {

 /// External relish storage configuration, enough for creating a client for that storage.
 #[derive(Debug, Clone)]
-pub struct RelishStorageConfig {
-    /// Limits the number of concurrent sync operations between pageserver and relish storage.
-    pub max_concurrent_sync: usize,
-    /// The storage connection configuration.
-    pub storage: RelishStorageKind,
-}
-
-/// A kind of a relish storage to connect to, with its connection configuration.
-#[derive(Debug, Clone)]
-pub enum RelishStorageKind {
-    /// Storage based on local file system.
-    /// Specify a root folder to place all stored relish data into.
+pub enum RelishStorageConfig {
+    /// Root folder to place all stored relish data into.
    LocalFs(PathBuf),
-    /// AWS S3 based storage, storing all relishes into the root
-    /// of the S3 bucket from the config.
    AwsS3(S3Config),
 }

 /// AWS S3 bucket coordinates and access credentials to manage the bucket contents (read and write).
 #[derive(Clone)]
 pub struct S3Config {
-    /// Name of the bucket to connect to.
    pub bucket_name: String,
-    /// The region where the bucket is located at.
    pub bucket_region: String,
-    /// "Login" to use when connecting to bucket.
-    /// Can be empty for cases like AWS k8s IAM
-    /// where we can allow certain pods to connect
-    /// to the bucket directly without any credentials.
    pub access_key_id: Option<String>,
-    /// "Password" to use when connecting to bucket.
    pub secret_access_key: Option<String>,
 }

--- a/pageserver/src/page_service.rs
+++ b/pageserver/src/page_service.rs
@@ -13,6 +13,7 @@
 use anyhow::{anyhow, bail, ensure, Result};
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 use lazy_static::lazy_static;
+use log::*;
 use regex::Regex;
 use std::net::TcpListener;
 use std::str;
@@ -20,12 +21,10 @@ use std::str::FromStr;
 use std::sync::Arc;
 use std::thread;
 use std::{io, net::TcpStream};
-use tracing::*;
 use zenith_metrics::{register_histogram_vec, HistogramVec};
 use zenith_utils::auth::{self, JwtAuth};
 use zenith_utils::auth::{Claims, Scope};
 use zenith_utils::lsn::Lsn;
-use zenith_utils::postgres_backend::is_socket_read_timed_out;
 use zenith_utils::postgres_backend::PostgresBackend;
 use zenith_utils::postgres_backend::{self, AuthType};
 use zenith_utils::pq_proto::{
@@ -188,32 +187,17 @@ pub fn thread_main(
    listener: TcpListener,
    auth_type: AuthType,
 ) -> anyhow::Result<()> {
-    let mut join_handles = Vec::new();
-
-    while !tenant_mgr::shutdown_requested() {
+    loop {
        let (socket, peer_addr) = listener.accept()?;
        debug!("accepted connection from {}", peer_addr);
        socket.set_nodelay(true).unwrap();
        let local_auth = auth.clone();
-
-        let handle = thread::Builder::new()
-            .name("serving Page Service thread".into())
-            .spawn(move || {
-                if let Err(err) = page_service_conn_main(conf, local_auth, socket, auth_type) {
-                    error!(%err, "page server thread exited with error");
-                }
-            })
-            .unwrap();
-
-        join_handles.push(handle);
+        thread::spawn(move || {
+            if let Err(err) = page_service_conn_main(conf, local_auth, socket, auth_type) {
+                error!("error: {}", err);
+            }
+        });
    }
-
-    debug!("page_service loop terminated. wait for connections to cancel");
-    for handle in join_handles.into_iter() {
-        handle.join().unwrap();
-    }
-
-    Ok(())
 }

 fn page_service_conn_main(
@@ -232,7 +216,7 @@ fn page_service_conn_main(
    }

    let mut conn_handler = PageServerHandler::new(conf, auth);
-    let pgbackend = PostgresBackend::new(socket, auth_type, None, true)?;
+    let pgbackend = PostgresBackend::new(socket, auth_type, None)?;
    pgbackend.run(&mut conn_handler)
 }

@@ -276,66 +260,48 @@ impl PageServerHandler {
        timelineid: ZTimelineId,
        tenantid: ZTenantId,
    ) -> anyhow::Result<()> {
-        let _enter = info_span!("pagestream", timeline = %timelineid, tenant = %tenantid).entered();
-
        // Check that the timeline exists
        let timeline = tenant_mgr::get_timeline_for_tenant(tenantid, timelineid)?;

        /* switch client to COPYBOTH */
        pgb.write_message(&BeMessage::CopyBothResponse)?;

-        while !tenant_mgr::shutdown_requested() {
-            match pgb.read_message() {
-                Ok(message) => {
-                    if let Some(message) = message {
-                        trace!("query: {:?}", message);
+        while let Some(message) = pgb.read_message()? {
+            trace!("query({:?}): {:?}", timelineid, message);

-                        let copy_data_bytes = match message {
-                            FeMessage::CopyData(bytes) => bytes,
-                            _ => continue,
-                        };
+            let copy_data_bytes = match message {
+                FeMessage::CopyData(bytes) => bytes,
+                _ => continue,
+            };

-                        let zenith_fe_msg = PagestreamFeMessage::parse(copy_data_bytes)?;
+            let zenith_fe_msg = PagestreamFeMessage::parse(copy_data_bytes)?;

-                        let response = match zenith_fe_msg {
-                            PagestreamFeMessage::Exists(req) => SMGR_QUERY_TIME
-                                .with_label_values(&["get_rel_exists"])
-                                .observe_closure_duration(|| {
-                                    self.handle_get_rel_exists_request(&*timeline, &req)
-                                }),
-                            PagestreamFeMessage::Nblocks(req) => SMGR_QUERY_TIME
-                                .with_label_values(&["get_rel_size"])
-                                .observe_closure_duration(|| {
-                                    self.handle_get_nblocks_request(&*timeline, &req)
-                                }),
-                            PagestreamFeMessage::GetPage(req) => SMGR_QUERY_TIME
-                                .with_label_values(&["get_page_at_lsn"])
-                                .observe_closure_duration(|| {
-                                    self.handle_get_page_at_lsn_request(&*timeline, &req)
-                                }),
-                        };
+            let response = match zenith_fe_msg {
+                PagestreamFeMessage::Exists(req) => SMGR_QUERY_TIME
+                    .with_label_values(&["get_rel_exists"])
+                    .observe_closure_duration(|| {
+                        self.handle_get_rel_exists_request(&*timeline, &req)
+                    }),
+                PagestreamFeMessage::Nblocks(req) => SMGR_QUERY_TIME
+                    .with_label_values(&["get_rel_size"])
+                    .observe_closure_duration(|| self.handle_get_nblocks_request(&*timeline, &req)),
+                PagestreamFeMessage::GetPage(req) => SMGR_QUERY_TIME
+                    .with_label_values(&["get_page_at_lsn"])
+                    .observe_closure_duration(|| {
+                        self.handle_get_page_at_lsn_request(&*timeline, &req)
+                    }),
+            };

-                        let response = response.unwrap_or_else(|e| {
-                            // print the all details to the log with {:#}, but for the client the
-                            // error message is enough
-                            error!("error reading relation or page version: {:#}", e);
-                            PagestreamBeMessage::Error(PagestreamErrorResponse {
-                                message: e.to_string(),
-                            })
-                        });
+            let response = response.unwrap_or_else(|e| {
+                error!("error reading relation or page version: {}", e);
+                PagestreamBeMessage::Error(PagestreamErrorResponse {
+                    message: e.to_string(),
+                })
+            });

-                        pgb.write_message(&BeMessage::CopyData(&response.serialize()))?;
-                    } else {
-                        break;
-                    }
-                }
-                Err(e) => {
-                    if !is_socket_read_timed_out(&e) {
-                        return Err(e);
-                    }
-                }
-            }
+            pgb.write_message(&BeMessage::CopyData(&response.serialize()))?;
        }
+
        Ok(())
    }

@@ -395,8 +361,6 @@ impl PageServerHandler {
        timeline: &dyn Timeline,
        req: &PagestreamExistsRequest,
    ) -> Result<PagestreamBeMessage> {
-        let _enter = info_span!("get_rel_exists", rel = %req.rel, req_lsn = %req.lsn).entered();
-
        let tag = RelishTag::Relation(req.rel);
        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest)?;

@@ -412,7 +376,6 @@ impl PageServerHandler {
        timeline: &dyn Timeline,
        req: &PagestreamNblocksRequest,
    ) -> Result<PagestreamBeMessage> {
-        let _enter = info_span!("get_nblocks", rel = %req.rel, req_lsn = %req.lsn).entered();
        let tag = RelishTag::Relation(req.rel);
        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest)?;

@@ -432,8 +395,6 @@ impl PageServerHandler {
        timeline: &dyn Timeline,
        req: &PagestreamGetPageRequest,
    ) -> Result<PagestreamBeMessage> {
-        let _enter = info_span!("get_page", rel = %req.rel, blkno = &req.blkno, req_lsn = %req.lsn)
-            .entered();
        let tag = RelishTag::Relation(req.rel);
        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest)?;

@@ -451,20 +412,17 @@ impl PageServerHandler {
        lsn: Option<Lsn>,
        tenantid: ZTenantId,
    ) -> anyhow::Result<()> {
-        let span = info_span!("basebackup", timeline = %timelineid, tenant = %tenantid, lsn = field::Empty);
-        let _enter = span.enter();
-
        // check that the timeline exists
        let timeline = tenant_mgr::get_timeline_for_tenant(tenantid, timelineid)?;

-        // switch client to COPYOUT
+        /* switch client to COPYOUT */
        pgb.write_message(&BeMessage::CopyOutResponse)?;
+        info!("sent CopyOut");

        /* Send a tarball of the latest layer on the timeline */
        {
            let mut writer = CopyDataSink { pgb };
            let mut basebackup = basebackup::Basebackup::new(&mut writer, &timeline, lsn)?;
-            span.record("lsn", &basebackup.lsn.to_string().as_str());
            basebackup.send_tarball()?;
        }
        pgb.write_message(&BeMessage::CopyDone)?;
@@ -569,6 +527,11 @@ impl postgres_backend::Handler for PageServerHandler {
                None
            };

+            info!(
+                "got basebackup command. tenantid=\"{}\" timelineid=\"{}\" lsn=\"{:#?}\"",
+                tenantid, timelineid, lsn
+            );
+
            // Check that the timeline exists
            self.handle_basebackup_request(pgb, timelineid, lsn, tenantid)?;
            pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
@@ -586,9 +549,6 @@ impl postgres_backend::Handler for PageServerHandler {

            self.check_permission(Some(tenantid))?;

-            let _enter =
-                info_span!("callmemaybe", timeline = %timelineid, tenant = %tenantid).entered();
-
            // Check that the timeline exists
            tenant_mgr::get_timeline_for_tenant(tenantid, timelineid)?;

@@ -611,9 +571,6 @@ impl postgres_backend::Handler for PageServerHandler {

            self.check_permission(Some(tenantid))?;

-            let _enter =
-                info_span!("branch_create", name = %branchname, tenant = %tenantid).entered();
-
            let branch =
                branches::create_branch(self.conf, &branchname, &startpoint_str, &tenantid)?;
            let branch = serde_json::to_vec(&branch)?;
@@ -697,14 +654,12 @@ impl postgres_backend::Handler for PageServerHandler {
                RowDescriptor::int8_col(b"layer_relfiles_needed_by_cutoff"),
                RowDescriptor::int8_col(b"layer_relfiles_needed_by_branches"),
                RowDescriptor::int8_col(b"layer_relfiles_not_updated"),
-                RowDescriptor::int8_col(b"layer_relfiles_needed_as_tombstone"),
                RowDescriptor::int8_col(b"layer_relfiles_removed"),
                RowDescriptor::int8_col(b"layer_relfiles_dropped"),
                RowDescriptor::int8_col(b"layer_nonrelfiles_total"),
                RowDescriptor::int8_col(b"layer_nonrelfiles_needed_by_cutoff"),
                RowDescriptor::int8_col(b"layer_nonrelfiles_needed_by_branches"),
                RowDescriptor::int8_col(b"layer_nonrelfiles_not_updated"),
-                RowDescriptor::int8_col(b"layer_nonrelfiles_needed_as_tombstone"),
                RowDescriptor::int8_col(b"layer_nonrelfiles_removed"),
                RowDescriptor::int8_col(b"layer_nonrelfiles_dropped"),
                RowDescriptor::int8_col(b"elapsed"),
@@ -724,12 +679,6 @@ impl postgres_backend::Handler for PageServerHandler {
                        .as_bytes(),
                ),
                Some(result.ondisk_relfiles_not_updated.to_string().as_bytes()),
-                Some(
-                    result
-                        .ondisk_relfiles_needed_as_tombstone
-                        .to_string()
-                        .as_bytes(),
-                ),
                Some(result.ondisk_relfiles_removed.to_string().as_bytes()),
                Some(result.ondisk_relfiles_dropped.to_string().as_bytes()),
                Some(result.ondisk_nonrelfiles_total.to_string().as_bytes()),
@@ -746,12 +695,6 @@ impl postgres_backend::Handler for PageServerHandler {
                        .as_bytes(),
                ),
                Some(result.ondisk_nonrelfiles_not_updated.to_string().as_bytes()),
-                Some(
-                    result
-                        .ondisk_nonrelfiles_needed_as_tombstone
-                        .to_string()
-                        .as_bytes(),
-                ),
                Some(result.ondisk_nonrelfiles_removed.to_string().as_bytes()),
                Some(result.ondisk_nonrelfiles_dropped.to_string().as_bytes()),
                Some(result.elapsed.as_millis().to_string().as_bytes()),
--- a/pageserver/src/relish_storage.rs
+++ b/pageserver/src/relish_storage.rs
@@ -8,41 +8,14 @@

 mod local_fs;
 mod rust_s3;
-/// A queue-based storage with the background machinery behind it to synchronize
-/// local page server layer files with external storage.
-mod synced_storage;
+/// A queue and the background machinery behind it to upload
+/// local page server layer files to external storage.
+pub mod storage_uploader;

-use std::{path::Path, thread};
+use std::path::Path;

 use anyhow::Context;

-pub use self::synced_storage::schedule_timeline_upload;
-use self::{local_fs::LocalFs, rust_s3::RustS3};
-use crate::{PageServerConf, RelishStorageKind};
-
-pub fn run_storage_sync_thread(
-    config: &'static PageServerConf,
-) -> anyhow::Result<Option<thread::JoinHandle<anyhow::Result<()>>>> {
-    match &config.relish_storage_config {
-        Some(relish_storage_config) => {
-            let max_concurrent_sync = relish_storage_config.max_concurrent_sync;
-            match &relish_storage_config.storage {
-                RelishStorageKind::LocalFs(root) => synced_storage::run_storage_sync_thread(
-                    config,
-                    LocalFs::new(root.clone())?,
-                    max_concurrent_sync,
-                ),
-                RelishStorageKind::AwsS3(s3_config) => synced_storage::run_storage_sync_thread(
-                    config,
-                    RustS3::new(s3_config)?,
-                    max_concurrent_sync,
-                ),
-            }
-        }
-        None => Ok(None),
-    }
-}
-
 /// Storage (potentially remote) API to manage its state.
 #[async_trait::async_trait]
 pub trait RelishStorage: Send + Sync {
@@ -55,21 +28,15 @@ pub trait RelishStorage: Send + Sync {

    async fn list_relishes(&self) -> anyhow::Result<Vec<Self::RelishStoragePath>>;

-    async fn download_relish<W: 'static + std::io::Write + Send>(
+    async fn download_relish(
        &self,
        from: &Self::RelishStoragePath,
-        // rust_s3 `get_object_stream` method requires `std::io::BufWriter` for some reason, not the async counterpart
-        // that forces us to consume and return the writer to satisfy the blocking operation async wrapper requirements
-        to: std::io::BufWriter<W>,
-    ) -> anyhow::Result<std::io::BufWriter<W>>;
+        to: &Path,
+    ) -> anyhow::Result<()>;

    async fn delete_relish(&self, path: &Self::RelishStoragePath) -> anyhow::Result<()>;

-    async fn upload_relish<R: tokio::io::AsyncRead + std::marker::Unpin + Send>(
-        &self,
-        from: &mut tokio::io::BufReader<R>,
-        to: &Self::RelishStoragePath,
-    ) -> anyhow::Result<()>;
+    async fn upload_relish(&self, from: &Path, to: &Self::RelishStoragePath) -> anyhow::Result<()>;
 }

 fn strip_workspace_prefix<'a>(
--- a/pageserver/src/relish_storage/local_fs.rs
+++ b/pageserver/src/relish_storage/local_fs.rs
@@ -9,13 +9,11 @@

 use std::{
    future::Future,
-    io::Write,
    path::{Path, PathBuf},
    pin::Pin,
 };

 use anyhow::{bail, Context};
-use tokio::{fs, io};

 use super::{strip_workspace_prefix, RelishStorage};

@@ -66,33 +64,16 @@ impl RelishStorage for LocalFs {
        Ok(get_all_files(&self.root).await?.into_iter().collect())
    }

-    async fn download_relish<W: 'static + std::io::Write + Send>(
+    async fn download_relish(
        &self,
        from: &Self::RelishStoragePath,
-        mut to: std::io::BufWriter<W>,
-    ) -> anyhow::Result<std::io::BufWriter<W>> {
+        to: &Path,
+    ) -> anyhow::Result<()> {
        let file_path = self.resolve_in_storage(from)?;
        if file_path.exists() && file_path.is_file() {
-            let updated_buffer = tokio::task::spawn_blocking(move || {
-                let mut source = std::io::BufReader::new(
-                    std::fs::OpenOptions::new()
-                        .read(true)
-                        .open(&file_path)
-                        .with_context(|| {
-                            format!(
-                                "Failed to open source file '{}' to use in the download",
-                                file_path.display()
-                            )
-                        })?,
-                );
-                std::io::copy(&mut source, &mut to)
-                    .context("Failed to download the relish file")?;
-                to.flush().context("Failed to flush the download buffer")?;
-                Ok::<_, anyhow::Error>(to)
-            })
-            .await
-            .context("Failed to spawn a blocking task")??;
-            Ok(updated_buffer)
+            create_target_directory(to).await?;
+            tokio::fs::copy(file_path, to).await?;
+            Ok(())
        } else {
            bail!(
                "File '{}' either does not exist or is not a file",
@@ -113,30 +94,18 @@ impl RelishStorage for LocalFs {
        }
    }

-    async fn upload_relish<R: io::AsyncRead + std::marker::Unpin + Send>(
-        &self,
-        from: &mut io::BufReader<R>,
-        to: &Self::RelishStoragePath,
-    ) -> anyhow::Result<()> {
+    async fn upload_relish(&self, from: &Path, to: &Self::RelishStoragePath) -> anyhow::Result<()> {
        let target_file_path = self.resolve_in_storage(to)?;
        create_target_directory(&target_file_path).await?;
-        let mut destination = io::BufWriter::new(
-            fs::OpenOptions::new()
-                .write(true)
-                .create(true)
-                .open(&target_file_path)
-                .await
-                .with_context(|| {
-                    format!(
-                        "Failed to open target fs destination at '{}'",
-                        target_file_path.display()
-                    )
-                })?,
-        );

-        io::copy_buf(from, &mut destination)
+        tokio::fs::copy(&from, &target_file_path)
            .await
-            .context("Failed to upload relish to local storage")?;
+            .with_context(|| {
+                format!(
+                    "Failed to upload relish '{}' to local storage",
+                    from.display(),
+                )
+            })?;
        Ok(())
    }
 }
--- a/pageserver/src/relish_storage/rust_s3.rs
+++ b/pageserver/src/relish_storage/rust_s3.rs
@@ -1,15 +1,13 @@
 //! A wrapper around AWS S3 client library `rust_s3` to be used a relish storage.

-use std::io::Write;
 use std::path::Path;

 use anyhow::Context;
 use s3::{bucket::Bucket, creds::Credentials, region::Region};

-use crate::{
-    relish_storage::{strip_workspace_prefix, RelishStorage},
-    S3Config,
-};
+use crate::{relish_storage::strip_workspace_prefix, S3Config};
+
+use super::RelishStorage;

 const S3_FILE_SEPARATOR: char = '/';

@@ -84,14 +82,18 @@ impl RelishStorage for RustS3 {
            .collect())
    }

-    async fn download_relish<W: 'static + std::io::Write + Send>(
+    async fn download_relish(
        &self,
        from: &Self::RelishStoragePath,
-        mut to: std::io::BufWriter<W>,
-    ) -> anyhow::Result<std::io::BufWriter<W>> {
+        to: &Path,
+    ) -> anyhow::Result<()> {
+        let mut target_file = std::fs::OpenOptions::new()
+            .write(true)
+            .open(to)
+            .with_context(|| format!("Failed to open target s3 destination at {}", to.display()))?;
        let code = self
            .bucket
-            .get_object_stream(from.key(), &mut to)
+            .get_object_stream(from.key(), &mut target_file)
            .await
            .with_context(|| format!("Failed to download s3 object with key {}", from.key()))?;
        if code != 200 {
@@ -100,12 +102,7 @@ impl RelishStorage for RustS3 {
                code
            ))
        } else {
-            tokio::task::spawn_blocking(move || {
-                to.flush().context("Failed to fluch the downoad buffer")?;
-                Ok::<_, anyhow::Error>(to)
-            })
-            .await
-            .context("Failed to joim the download buffer flush task")?
+            Ok(())
        }
    }

@@ -126,14 +123,12 @@ impl RelishStorage for RustS3 {
        }
    }

-    async fn upload_relish<R: tokio::io::AsyncRead + std::marker::Unpin + Send>(
-        &self,
-        from: &mut tokio::io::BufReader<R>,
-        to: &Self::RelishStoragePath,
-    ) -> anyhow::Result<()> {
+    async fn upload_relish(&self, from: &Path, to: &Self::RelishStoragePath) -> anyhow::Result<()> {
+        let mut local_file = tokio::fs::OpenOptions::new().read(true).open(from).await?;
+
        let code = self
            .bucket
-            .put_object_stream(from, to.key())
+            .put_object_stream(&mut local_file, to.key())
            .await
            .with_context(|| format!("Failed to create s3 object with key {}", to.key()))?;
        if code != 200 {
--- a/pageserver/src/relish_storage/storage_uploader.rs
+++ b/pageserver/src/relish_storage/storage_uploader.rs
@@ -0,0 +1,116 @@
+use std::{
+    collections::VecDeque,
+    path::{Path, PathBuf},
+    sync::{Arc, Mutex},
+    thread,
+};
+
+use zenith_utils::zid::ZTimelineId;
+
+use crate::{relish_storage::RelishStorage, RelishStorageConfig};
+
+use super::{local_fs::LocalFs, rust_s3::RustS3};
+
+pub struct QueueBasedRelishUploader {
+    upload_queue: Arc<Mutex<VecDeque<(ZTimelineId, PathBuf)>>>,
+}
+
+impl QueueBasedRelishUploader {
+    pub fn new(
+        config: &RelishStorageConfig,
+        page_server_workdir: &'static Path,
+    ) -> anyhow::Result<Self> {
+        let upload_queue = Arc::new(Mutex::new(VecDeque::new()));
+        let _handle = match config {
+            RelishStorageConfig::LocalFs(root) => {
+                let relish_storage = LocalFs::new(root.clone())?;
+                create_upload_thread(
+                    Arc::clone(&upload_queue),
+                    relish_storage,
+                    page_server_workdir,
+                )?
+            }
+            RelishStorageConfig::AwsS3(s3_config) => {
+                let relish_storage = RustS3::new(s3_config)?;
+                create_upload_thread(
+                    Arc::clone(&upload_queue),
+                    relish_storage,
+                    page_server_workdir,
+                )?
+            }
+        };
+
+        Ok(Self { upload_queue })
+    }
+
+    pub fn schedule_upload(&self, timeline_id: ZTimelineId, relish_path: PathBuf) {
+        self.upload_queue
+            .lock()
+            .unwrap()
+            .push_back((timeline_id, relish_path))
+    }
+}
+
+fn create_upload_thread<P, S: 'static + RelishStorage<RelishStoragePath = P>>(
+    upload_queue: Arc<Mutex<VecDeque<(ZTimelineId, PathBuf)>>>,
+    relish_storage: S,
+    page_server_workdir: &'static Path,
+) -> std::io::Result<thread::JoinHandle<()>> {
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()?;
+    thread::Builder::new()
+        .name("Queue based relish uploader".to_string())
+        .spawn(move || loop {
+            runtime.block_on(async {
+                upload_loop_step(&upload_queue, &relish_storage, page_server_workdir).await;
+            })
+        })
+}
+
+async fn upload_loop_step<P, S: 'static + RelishStorage<RelishStoragePath = P>>(
+    upload_queue: &Mutex<VecDeque<(ZTimelineId, PathBuf)>>,
+    relish_storage: &S,
+    page_server_workdir: &Path,
+) {
+    let mut queue_accessor = upload_queue.lock().unwrap();
+    log::debug!("current upload queue length: {}", queue_accessor.len());
+    let next_upload = queue_accessor.pop_front();
+    drop(queue_accessor);
+
+    let (relish_timeline_id, relish_local_path) = match next_upload {
+        Some(data) => data,
+        None => {
+            // Don't spin and allow others to use the queue.
+            // In future, could be improved to be more clever about delays depending on relish upload stats
+            thread::sleep(std::time::Duration::from_secs(1));
+            return;
+        }
+    };
+
+    if let Err(e) = upload_relish(relish_storage, page_server_workdir, &relish_local_path).await {
+        log::error!(
+            "Failed to upload relish '{}' for timeline {}, reason: {}",
+            relish_local_path.display(),
+            relish_timeline_id,
+            e
+        );
+        upload_queue
+            .lock()
+            .unwrap()
+            .push_back((relish_timeline_id, relish_local_path))
+    } else {
+        log::debug!("Relish successfully uploaded");
+    }
+}
+
+async fn upload_relish<P, S: RelishStorage<RelishStoragePath = P>>(
+    relish_storage: &S,
+    page_server_workdir: &Path,
+    relish_local_path: &Path,
+) -> anyhow::Result<()> {
+    let destination = S::derive_destination(page_server_workdir, relish_local_path)?;
+    relish_storage
+        .upload_relish(relish_local_path, &destination)
+        .await
+}
--- a/pageserver/src/relish_storage/synced_storage.rs
+++ b/pageserver/src/relish_storage/synced_storage.rs
@@ -1,57 +0,0 @@
-use std::time::Duration;
-use std::{collections::BinaryHeap, sync::Mutex, thread};
-
-use crate::tenant_mgr;
-use crate::{relish_storage::RelishStorage, PageServerConf};
-
-lazy_static::lazy_static! {
-    static ref UPLOAD_QUEUE: Mutex<BinaryHeap<SyncTask>> = Mutex::new(BinaryHeap::new());
-}
-
-pub fn schedule_timeline_upload(_local_timeline: ()) {
-    // UPLOAD_QUEUE
-    //     .lock()
-    //     .unwrap()
-    //     .push(SyncTask::Upload(local_timeline))
-}
-
-#[derive(Debug, PartialEq, Eq, PartialOrd, Ord)]
-enum SyncTask {}
-
-pub fn run_storage_sync_thread<
-    P: std::fmt::Debug,
-    S: 'static + RelishStorage<RelishStoragePath = P>,
->(
-    config: &'static PageServerConf,
-    relish_storage: S,
-    max_concurrent_sync: usize,
-) -> anyhow::Result<Option<thread::JoinHandle<anyhow::Result<()>>>> {
-    let runtime = tokio::runtime::Builder::new_current_thread()
-        .enable_all()
-        .build()?;
-
-    let handle = thread::Builder::new()
-        .name("Queue based relish storage sync".to_string())
-        .spawn(move || {
-            while !tenant_mgr::shutdown_requested() {
-                let mut queue_accessor = UPLOAD_QUEUE.lock().unwrap();
-                log::debug!("Upload queue length: {}", queue_accessor.len());
-                let next_task = queue_accessor.pop();
-                drop(queue_accessor);
-                match next_task {
-                    Some(task) => runtime.block_on(async {
-                        // suppress warnings
-                        let _ = (config, task, &relish_storage, max_concurrent_sync);
-                        todo!("omitted for brevity")
-                    }),
-                    None => {
-                        thread::sleep(Duration::from_secs(1));
-                        continue;
-                    }
-                }
-            }
-            log::debug!("Queue based relish storage sync thread shut down");
-            Ok(())
-        })?;
-    Ok(Some(handle))
-}
--- a/pageserver/src/repository.rs
+++ b/pageserver/src/repository.rs
@@ -3,7 +3,7 @@ use anyhow::Result;
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 use serde::{Deserialize, Serialize};
 use std::collections::HashSet;
-use std::ops::{AddAssign, Deref};
+use std::ops::AddAssign;
 use std::sync::Arc;
 use std::time::Duration;
 use zenith_utils::lsn::{Lsn, RecordLsn};
@@ -13,8 +13,6 @@ use zenith_utils::zid::ZTimelineId;
 /// A repository corresponds to one .zenith directory. One repository holds multiple
 /// timelines, forked off from the same initial call to 'initdb'.
 pub trait Repository: Send + Sync {
-    fn shutdown(&self) -> Result<()>;
-
    /// Get Timeline handle for given zenith timeline ID.
    fn get_timeline(&self, timelineid: ZTimelineId) -> Result<Arc<dyn Timeline>>;

@@ -30,14 +28,18 @@ pub trait Repository: Send + Sync {
    ///
    /// 'timelineid' specifies the timeline to GC, or None for all.
    /// `horizon` specifies delta from last lsn to preserve all object versions (pitr interval).
-    /// `checkpoint_before_gc` parameter is used to force compaction of storage before CG
-    /// to make tests more deterministic.
-    /// TODO Do we still need it or we can call checkpoint explicitly in tests where needed?
+    /// `compact` parameter is used to force compaction of storage.
+    /// some storage implementation are based on lsm tree and require periodic merge (compaction).
+    /// usually storage implementation determines itself when compaction should be performed.
+    /// but for gc tests it way be useful to force compaction just after completion of gc iteration
+    /// to make sure that all detected garbage is removed.
+    /// so right now `compact` is set to true when gc explicitly requested through page srver api,
+    /// and is st to false in gc threads which infinitely repeats gc iterations in loop.
    fn gc_iteration(
        &self,
        timelineid: Option<ZTimelineId>,
        horizon: u64,
-        checkpoint_before_gc: bool,
+        compact: bool,
    ) -> Result<GcResult>;
 }

@@ -50,7 +52,6 @@ pub struct GcResult {
    pub ondisk_relfiles_needed_by_cutoff: u64,
    pub ondisk_relfiles_needed_by_branches: u64,
    pub ondisk_relfiles_not_updated: u64,
-    pub ondisk_relfiles_needed_as_tombstone: u64,
    pub ondisk_relfiles_removed: u64, // # of layer files removed because they have been made obsolete by newer ondisk files.
    pub ondisk_relfiles_dropped: u64, // # of layer files removed because the relation was dropped

@@ -58,7 +59,6 @@ pub struct GcResult {
    pub ondisk_nonrelfiles_needed_by_cutoff: u64,
    pub ondisk_nonrelfiles_needed_by_branches: u64,
    pub ondisk_nonrelfiles_not_updated: u64,
-    pub ondisk_nonrelfiles_needed_as_tombstone: u64,
    pub ondisk_nonrelfiles_removed: u64, // # of layer files removed because they have been made obsolete by newer ondisk files.
    pub ondisk_nonrelfiles_dropped: u64, // # of layer files removed because the relation was dropped

@@ -71,7 +71,6 @@ impl AddAssign for GcResult {
        self.ondisk_relfiles_needed_by_cutoff += other.ondisk_relfiles_needed_by_cutoff;
        self.ondisk_relfiles_needed_by_branches += other.ondisk_relfiles_needed_by_branches;
        self.ondisk_relfiles_not_updated += other.ondisk_relfiles_not_updated;
-        self.ondisk_relfiles_needed_as_tombstone += other.ondisk_relfiles_needed_as_tombstone;
        self.ondisk_relfiles_removed += other.ondisk_relfiles_removed;
        self.ondisk_relfiles_dropped += other.ondisk_relfiles_dropped;

@@ -79,7 +78,6 @@ impl AddAssign for GcResult {
        self.ondisk_nonrelfiles_needed_by_cutoff += other.ondisk_nonrelfiles_needed_by_cutoff;
        self.ondisk_nonrelfiles_needed_by_branches += other.ondisk_nonrelfiles_needed_by_branches;
        self.ondisk_nonrelfiles_not_updated += other.ondisk_nonrelfiles_not_updated;
-        self.ondisk_nonrelfiles_needed_as_tombstone += other.ondisk_nonrelfiles_needed_as_tombstone;
        self.ondisk_nonrelfiles_removed += other.ondisk_nonrelfiles_removed;
        self.ondisk_nonrelfiles_dropped += other.ondisk_nonrelfiles_dropped;

@@ -125,6 +123,26 @@ pub trait Timeline: Send + Sync {
    // These are called by the WAL receiver to digest WAL records.
    //------------------------------------------------------------------------------

+    /// Put a new page version that can be constructed from a WAL record
+    ///
+    /// This will implicitly extend the relation, if the page is beyond the
+    /// current end-of-file.
+    fn put_wal_record(&self, tag: RelishTag, blknum: u32, rec: WALRecord) -> Result<()>;
+
+    /// Like put_wal_record, but with ready-made image of the page.
+    fn put_page_image(&self, tag: RelishTag, blknum: u32, lsn: Lsn, img: Bytes) -> Result<()>;
+
+    /// Truncate relation
+    fn put_truncation(&self, rel: RelishTag, lsn: Lsn, nblocks: u32) -> Result<()>;
+
+    /// This method is used for marking dropped relations and truncated SLRU files and aborted two phase records
+    fn drop_relish(&self, tag: RelishTag, lsn: Lsn) -> Result<()>;
+
+    /// Track end of the latest digested WAL record.
+    ///
+    /// Advance requires aligned LSN as an argument and would wake wait_lsn() callers.
+    /// Previous last record LSN is stored alongside the latest and can be read.
+    fn advance_last_record_lsn(&self, lsn: Lsn);
    /// Atomically get both last and prev.
    fn get_last_record_rlsn(&self) -> RecordLsn;
    /// Get last or prev record separately. Same as get_last_record_rlsn().last/prev.
@@ -132,9 +150,6 @@ pub trait Timeline: Send + Sync {
    fn get_prev_record_lsn(&self) -> Lsn;
    fn get_start_lsn(&self) -> Lsn;

-    /// Mutate the timeline with a [`TimelineWriter`].
-    fn writer<'a>(&'a self) -> Box<dyn TimelineWriter + 'a>;
-
    ///
    /// Flush to disk all data that was written with the put_* functions
    ///
@@ -153,35 +168,9 @@ pub trait Timeline: Send + Sync {
    fn get_current_logical_size_non_incremental(&self, lsn: Lsn) -> Result<usize>;
 }

-/// Various functions to mutate the timeline.
-// TODO Currently, Deref is used to allow easy access to read methods from this trait.
-// This is probably considered a bad practice in Rust and should be fixed eventually,
-// but will cause large code changes.
-pub trait TimelineWriter: Deref<Target = dyn Timeline> {
-    /// Put a new page version that can be constructed from a WAL record
-    ///
-    /// This will implicitly extend the relation, if the page is beyond the
-    /// current end-of-file.
-    fn put_wal_record(&self, lsn: Lsn, tag: RelishTag, blknum: u32, rec: WALRecord) -> Result<()>;
-
-    /// Like put_wal_record, but with ready-made image of the page.
-    fn put_page_image(&self, tag: RelishTag, blknum: u32, lsn: Lsn, img: Bytes) -> Result<()>;
-
-    /// Truncate relation
-    fn put_truncation(&self, rel: RelishTag, lsn: Lsn, nblocks: u32) -> Result<()>;
-
-    /// This method is used for marking dropped relations and truncated SLRU files and aborted two phase records
-    fn drop_relish(&self, tag: RelishTag, lsn: Lsn) -> Result<()>;
-
-    /// Track end of the latest digested WAL record.
-    ///
-    /// Advance requires aligned LSN as an argument and would wake wait_lsn() callers.
-    /// Previous last record LSN is stored alongside the latest and can be read.
-    fn advance_last_record_lsn(&self, lsn: Lsn);
-}
-
 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
 pub struct WALRecord {
+    pub lsn: Lsn, // LSN at the *end* of the record
    pub will_init: bool,
    pub rec: Bytes,
    // Remember the offset of main_data in rec,
@@ -192,19 +181,22 @@ pub struct WALRecord {

 impl WALRecord {
    pub fn pack(&self, buf: &mut BytesMut) {
+        buf.put_u64(self.lsn.0);
        buf.put_u8(self.will_init as u8);
        buf.put_u32(self.main_data_offset);
        buf.put_u32(self.rec.len() as u32);
        buf.put_slice(&self.rec[..]);
    }
    pub fn unpack(buf: &mut Bytes) -> WALRecord {
+        let lsn = Lsn::from(buf.get_u64());
        let will_init = buf.get_u8() != 0;
        let main_data_offset = buf.get_u32();
-        let rec_len = buf.get_u32() as usize;
-        let rec = buf.split_to(rec_len);
+        let mut dst = vec![0u8; buf.get_u32() as usize];
+        buf.copy_to_slice(&mut dst);
        WALRecord {
+            lsn,
            will_init,
-            rec,
+            rec: Bytes::from(dst),
            main_data_offset,
        }
    }
@@ -217,21 +209,15 @@ impl WALRecord {
 #[cfg(test)]
 mod tests {
    use super::*;
-    use crate::layered_repository::{LayeredRepository, METADATA_FILE_NAME};
+    use crate::layered_repository::LayeredRepository;
    use crate::walredo::{WalRedoError, WalRedoManager};
    use crate::PageServerConf;
-    use hex_literal::hex;
    use postgres_ffi::pg_constants;
    use postgres_ffi::xlog_utils::SIZEOF_CHECKPOINT;
    use std::fs;
-    use std::path::PathBuf;
+    use std::str::FromStr;
    use zenith_utils::zid::ZTenantId;

-    const TIMELINE_ID: ZTimelineId =
-        ZTimelineId::from_array(hex!("11223344556677881122334455667788"));
-    const NEW_TIMELINE_ID: ZTimelineId =
-        ZTimelineId::from_array(hex!("AA223344556677881122334455667788"));
-
    /// Arbitrary relation tag, for testing.
    const TESTREL_A: RelishTag = RelishTag::Relation(RelTag {
        spcnode: 0,
@@ -267,62 +253,47 @@ mod tests {
    static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; 8192]);
    static ZERO_CHECKPOINT: Bytes = Bytes::from_static(&[0u8; SIZEOF_CHECKPOINT]);

-    struct RepoHarness {
-        conf: &'static PageServerConf,
-        tenant_id: ZTenantId,
-    }
+    fn get_test_repo(test_name: &str) -> Result<Box<dyn Repository>> {
+        let repo_dir = PageServerConf::test_repo_dir(test_name);
+        let _ = fs::remove_dir_all(&repo_dir);
+        fs::create_dir_all(&repo_dir)?;
+        fs::create_dir_all(&repo_dir.join("timelines"))?;

-    impl RepoHarness {
-        fn create(test_name: &'static str) -> Result<Self> {
-            let repo_dir = PageServerConf::test_repo_dir(test_name);
-            let _ = fs::remove_dir_all(&repo_dir);
-            fs::create_dir_all(&repo_dir)?;
-            fs::create_dir_all(&repo_dir.join("timelines"))?;
+        let conf = PageServerConf::dummy_conf(repo_dir);
+        // Make a static copy of the config. This can never be free'd, but that's
+        // OK in a test.
+        let conf: &'static PageServerConf = Box::leak(Box::new(conf));
+        let tenantid = ZTenantId::generate();
+        fs::create_dir_all(conf.tenant_path(&tenantid)).unwrap();

-            let conf = PageServerConf::dummy_conf(repo_dir);
-            // Make a static copy of the config. This can never be free'd, but that's
-            // OK in a test.
-            let conf: &'static PageServerConf = Box::leak(Box::new(conf));
+        let walredo_mgr = TestRedoManager {};

-            let tenant_id = ZTenantId::generate();
-            fs::create_dir_all(conf.tenant_path(&tenant_id))?;
+        let repo = Box::new(LayeredRepository::new(
+            conf,
+            Arc::new(walredo_mgr),
+            tenantid,
+        ));

-            Ok(Self { conf, tenant_id })
-        }
-
-        fn load(&self) -> Box<dyn Repository> {
-            let walredo_mgr = Arc::new(TestRedoManager);
-
-            Box::new(LayeredRepository::new(
-                self.conf,
-                walredo_mgr,
-                self.tenant_id,
-                false,
-            ))
-        }
-
-        fn timeline_path(&self, timeline_id: &ZTimelineId) -> PathBuf {
-            self.conf.timeline_path(timeline_id, &self.tenant_id)
-        }
+        Ok(repo)
    }

    #[test]
    fn test_relsize() -> Result<()> {
-        let repo = RepoHarness::create("test_relsize")?.load();
+        let repo = get_test_repo("test_relsize")?;
        // get_timeline() with non-existent timeline id should fail
        //repo.get_timeline("11223344556677881122334455667788");

        // Create timeline to work on
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
-        let writer = tline.writer();
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;

-        writer.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
-        writer.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
-        writer.put_page_image(TESTREL_A, 0, Lsn(0x30), TEST_IMG("foo blk 0 at 3"))?;
-        writer.put_page_image(TESTREL_A, 1, Lsn(0x40), TEST_IMG("foo blk 1 at 4"))?;
-        writer.put_page_image(TESTREL_A, 2, Lsn(0x50), TEST_IMG("foo blk 2 at 5"))?;
+        tline.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
+        tline.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
+        tline.put_page_image(TESTREL_A, 0, Lsn(0x30), TEST_IMG("foo blk 0 at 3"))?;
+        tline.put_page_image(TESTREL_A, 1, Lsn(0x40), TEST_IMG("foo blk 1 at 4"))?;
+        tline.put_page_image(TESTREL_A, 2, Lsn(0x50), TEST_IMG("foo blk 2 at 5"))?;

-        writer.advance_last_record_lsn(Lsn(0x50));
+        tline.advance_last_record_lsn(Lsn(0x50));

        assert_current_logical_size(&tline, Lsn(0x50));

@@ -368,8 +339,8 @@ mod tests {
        );

        // Truncate last block
-        writer.put_truncation(TESTREL_A, Lsn(0x60), 2)?;
-        writer.advance_last_record_lsn(Lsn(0x60));
+        tline.put_truncation(TESTREL_A, Lsn(0x60), 2)?;
+        tline.advance_last_record_lsn(Lsn(0x60));
        assert_current_logical_size(&tline, Lsn(0x60));

        // Check reported size and contents after truncation
@@ -391,13 +362,13 @@ mod tests {
        );

        // Truncate to zero length
-        writer.put_truncation(TESTREL_A, Lsn(0x68), 0)?;
-        writer.advance_last_record_lsn(Lsn(0x68));
+        tline.put_truncation(TESTREL_A, Lsn(0x68), 0)?;
+        tline.advance_last_record_lsn(Lsn(0x68));
        assert_eq!(tline.get_relish_size(TESTREL_A, Lsn(0x68))?.unwrap(), 0);

        // Extend from 0 to 2 blocks, leaving a gap
-        writer.put_page_image(TESTREL_A, 1, Lsn(0x70), TEST_IMG("foo blk 1"))?;
-        writer.advance_last_record_lsn(Lsn(0x70));
+        tline.put_page_image(TESTREL_A, 1, Lsn(0x70), TEST_IMG("foo blk 1"))?;
+        tline.advance_last_record_lsn(Lsn(0x70));
        assert_eq!(tline.get_relish_size(TESTREL_A, Lsn(0x70))?.unwrap(), 2);
        assert_eq!(tline.get_page_at_lsn(TESTREL_A, 0, Lsn(0x70))?, ZERO_PAGE);
        assert_eq!(
@@ -424,159 +395,21 @@ mod tests {
        Ok(())
    }

-    // Test what happens if we dropped a relation
-    // and then created it again within the same layer.
-    #[test]
-    fn test_drop_extend() -> Result<()> {
-        let repo = RepoHarness::create("test_drop_extend")?.load();
-
-        // Create timeline to work on
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
-        let writer = tline.writer();
-
-        writer.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
-        writer.advance_last_record_lsn(Lsn(0x20));
-
-        // Check that rel exists and size is correct
-        assert_eq!(tline.get_rel_exists(TESTREL_A, Lsn(0x20))?, true);
-        assert_eq!(tline.get_relish_size(TESTREL_A, Lsn(0x20))?.unwrap(), 1);
-
-        // Drop relish
-        writer.drop_relish(TESTREL_A, Lsn(0x30))?;
-        writer.advance_last_record_lsn(Lsn(0x30));
-
-        // Check that rel is not visible anymore
-        assert_eq!(tline.get_rel_exists(TESTREL_A, Lsn(0x30))?, false);
-        assert!(tline.get_relish_size(TESTREL_A, Lsn(0x30))?.is_none());
-
-        // Extend it again
-        writer.put_page_image(TESTREL_A, 0, Lsn(0x40), TEST_IMG("foo blk 0 at 4"))?;
-        writer.advance_last_record_lsn(Lsn(0x40));
-
-        // Check that rel exists and size is correct
-        assert_eq!(tline.get_rel_exists(TESTREL_A, Lsn(0x40))?, true);
-        assert_eq!(tline.get_relish_size(TESTREL_A, Lsn(0x40))?.unwrap(), 1);
-
-        Ok(())
-    }
-
-    // Test what happens if we truncated a relation
-    // so that one of its segments was dropped
-    // and then extended it again within the same layer.
-    #[test]
-    fn test_truncate_extend() -> Result<()> {
-        let repo = RepoHarness::create("test_truncate_extend")?.load();
-
-        // Create timeline to work on
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
-        let writer = tline.writer();
-
-        //from storage_layer.rs
-        const RELISH_SEG_SIZE: u32 = 10 * 1024 * 1024 / 8192;
-        let relsize = RELISH_SEG_SIZE * 2;
-
-        // Create relation with relsize blocks
-        for blkno in 0..relsize {
-            let lsn = Lsn(0x20);
-            let data = format!("foo blk {} at {}", blkno, lsn);
-            writer.put_page_image(TESTREL_A, blkno, lsn, TEST_IMG(&data))?;
-        }
-
-        writer.advance_last_record_lsn(Lsn(0x20));
-
-        // The relation was created at LSN 2, not visible at LSN 1 yet.
-        assert_eq!(tline.get_rel_exists(TESTREL_A, Lsn(0x10))?, false);
-        assert!(tline.get_relish_size(TESTREL_A, Lsn(0x10))?.is_none());
-
-        assert_eq!(tline.get_rel_exists(TESTREL_A, Lsn(0x20))?, true);
-        assert_eq!(
-            tline.get_relish_size(TESTREL_A, Lsn(0x20))?.unwrap(),
-            relsize
-        );
-
-        // Check relation content
-        for blkno in 0..relsize {
-            let lsn = Lsn(0x20);
-            let data = format!("foo blk {} at {}", blkno, lsn);
-            assert_eq!(
-                tline.get_page_at_lsn(TESTREL_A, blkno, lsn)?,
-                TEST_IMG(&data)
-            );
-        }
-
-        // Truncate relation so that second segment was dropped
-        // - only leave one page
-        writer.put_truncation(TESTREL_A, Lsn(0x60), 1)?;
-        writer.advance_last_record_lsn(Lsn(0x60));
-
-        // Check reported size and contents after truncation
-        assert_eq!(tline.get_relish_size(TESTREL_A, Lsn(0x60))?.unwrap(), 1);
-
-        for blkno in 0..1 {
-            let lsn = Lsn(0x20);
-            let data = format!("foo blk {} at {}", blkno, lsn);
-            assert_eq!(
-                tline.get_page_at_lsn(TESTREL_A, blkno, Lsn(0x60))?,
-                TEST_IMG(&data)
-            );
-        }
-
-        // should still see all blocks with older LSN
-        assert_eq!(
-            tline.get_relish_size(TESTREL_A, Lsn(0x50))?.unwrap(),
-            relsize
-        );
-        for blkno in 0..relsize {
-            let lsn = Lsn(0x20);
-            let data = format!("foo blk {} at {}", blkno, lsn);
-            assert_eq!(
-                tline.get_page_at_lsn(TESTREL_A, blkno, Lsn(0x50))?,
-                TEST_IMG(&data)
-            );
-        }
-
-        // Extend relation again.
-        // Add enough blocks to create second segment
-        for blkno in 0..relsize {
-            let lsn = Lsn(0x80);
-            let data = format!("foo blk {} at {}", blkno, lsn);
-            writer.put_page_image(TESTREL_A, blkno, lsn, TEST_IMG(&data))?;
-        }
-        writer.advance_last_record_lsn(Lsn(0x80));
-
-        assert_eq!(tline.get_rel_exists(TESTREL_A, Lsn(0x80))?, true);
-        assert_eq!(
-            tline.get_relish_size(TESTREL_A, Lsn(0x80))?.unwrap(),
-            relsize
-        );
-        // Check relation content
-        for blkno in 0..relsize {
-            let lsn = Lsn(0x80);
-            let data = format!("foo blk {} at {}", blkno, lsn);
-            assert_eq!(
-                tline.get_page_at_lsn(TESTREL_A, blkno, Lsn(0x80))?,
-                TEST_IMG(&data)
-            );
-        }
-
-        Ok(())
-    }
-
    /// Test get_relsize() and truncation with a file larger than 1 GB, so that it's
    /// split into multiple 1 GB segments in Postgres.
    #[test]
    fn test_large_rel() -> Result<()> {
-        let repo = RepoHarness::create("test_large_rel")?.load();
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
-        let writer = tline.writer();
+        let repo = get_test_repo("test_large_rel")?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;

        let mut lsn = 0x10;
        for blknum in 0..pg_constants::RELSEG_SIZE + 1 {
            let img = TEST_IMG(&format!("foo blk {} at {}", blknum, Lsn(lsn)));
            lsn += 0x10;
-            writer.put_page_image(TESTREL_A, blknum as u32, Lsn(lsn), img)?;
+            tline.put_page_image(TESTREL_A, blknum as u32, Lsn(lsn), img)?;
        }
-        writer.advance_last_record_lsn(Lsn(lsn));
+        tline.advance_last_record_lsn(Lsn(lsn));

        assert_current_logical_size(&tline, Lsn(lsn));

@@ -587,8 +420,8 @@ mod tests {

        // Truncate one block
        lsn += 0x10;
-        writer.put_truncation(TESTREL_A, Lsn(lsn), pg_constants::RELSEG_SIZE)?;
-        writer.advance_last_record_lsn(Lsn(lsn));
+        tline.put_truncation(TESTREL_A, Lsn(lsn), pg_constants::RELSEG_SIZE)?;
+        tline.advance_last_record_lsn(Lsn(lsn));
        assert_eq!(
            tline.get_relish_size(TESTREL_A, Lsn(lsn))?.unwrap(),
            pg_constants::RELSEG_SIZE
@@ -597,8 +430,8 @@ mod tests {

        // Truncate another block
        lsn += 0x10;
-        writer.put_truncation(TESTREL_A, Lsn(lsn), pg_constants::RELSEG_SIZE - 1)?;
-        writer.advance_last_record_lsn(Lsn(lsn));
+        tline.put_truncation(TESTREL_A, Lsn(lsn), pg_constants::RELSEG_SIZE - 1)?;
+        tline.advance_last_record_lsn(Lsn(lsn));
        assert_eq!(
            tline.get_relish_size(TESTREL_A, Lsn(lsn))?.unwrap(),
            pg_constants::RELSEG_SIZE - 1
@@ -610,8 +443,8 @@ mod tests {
        let mut size: i32 = 3000;
        while size >= 0 {
            lsn += 0x10;
-            writer.put_truncation(TESTREL_A, Lsn(lsn), size as u32)?;
-            writer.advance_last_record_lsn(Lsn(lsn));
+            tline.put_truncation(TESTREL_A, Lsn(lsn), size as u32)?;
+            tline.advance_last_record_lsn(Lsn(lsn));
            assert_eq!(
                tline.get_relish_size(TESTREL_A, Lsn(lsn))?.unwrap(),
                size as u32
@@ -628,20 +461,23 @@ mod tests {
    /// Test list_rels() function, with branches and dropped relations
    ///
    #[test]
+    // FIXME: The last assertion in this test is currently failing, see
+    // https://github.com/zenithdb/zenith/issues/502. Ignore the failure until that's fixed.
+    #[ignore]
    fn test_list_rels_drop() -> Result<()> {
-        let repo = RepoHarness::create("test_list_rels_drop")?.load();
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
-        let writer = tline.writer();
+        let repo = get_test_repo("test_list_rels_drop")?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;
        const TESTDB: u32 = 111;

        // Import initial dummy checkpoint record, otherwise the get_timeline() call
        // after branching fails below
-        writer.put_page_image(RelishTag::Checkpoint, 0, Lsn(0x10), ZERO_CHECKPOINT.clone())?;
+        tline.put_page_image(RelishTag::Checkpoint, 0, Lsn(0x10), ZERO_CHECKPOINT.clone())?;

        // Create a relation on the timeline
-        writer.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
+        tline.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;

-        writer.advance_last_record_lsn(Lsn(0x30));
+        tline.advance_last_record_lsn(Lsn(0x30));

        // Check that list_rels() lists it after LSN 2, but no before it
        assert!(!tline.list_rels(0, TESTDB, Lsn(0x10))?.contains(&TESTREL_A));
@@ -649,19 +485,17 @@ mod tests {
        assert!(tline.list_rels(0, TESTDB, Lsn(0x30))?.contains(&TESTREL_A));

        // Create a branch, check that the relation is visible there
-        repo.branch_timeline(TIMELINE_ID, NEW_TIMELINE_ID, Lsn(0x30))?;
-        let newtline = repo.get_timeline(NEW_TIMELINE_ID)?;
-        let new_writer = newtline.writer();
+        let newtimelineid = ZTimelineId::from_str("AA223344556677881122334455667788").unwrap();
+        repo.branch_timeline(timelineid, newtimelineid, Lsn(0x30))?;
+        let newtline = repo.get_timeline(newtimelineid)?;

        assert!(newtline
            .list_rels(0, TESTDB, Lsn(0x30))?
            .contains(&TESTREL_A));

        // Drop it on the branch
-        new_writer.drop_relish(TESTREL_A, Lsn(0x40))?;
-        new_writer.advance_last_record_lsn(Lsn(0x40));
-
-        drop(new_writer);
+        newtline.drop_relish(TESTREL_A, Lsn(0x40))?;
+        newtline.advance_last_record_lsn(Lsn(0x40));

        // Check that it's no longer listed on the branch after the point where it was dropped
        assert!(newtline
@@ -673,8 +507,9 @@ mod tests {

        // Run checkpoint and garbage collection and check that it's still not visible
        newtline.checkpoint()?;
-        repo.gc_iteration(Some(NEW_TIMELINE_ID), 0, true)?;
+        repo.gc_iteration(Some(newtimelineid), 0, true)?;

+        // FIXME: this is currently failing
        assert!(!newtline
            .list_rels(0, TESTDB, Lsn(0x40))?
            .contains(&TESTREL_A));
@@ -687,32 +522,32 @@ mod tests {
    ///
    #[test]
    fn test_branch() -> Result<()> {
-        let repo = RepoHarness::create("test_branch")?.load();
-        let tline = repo.create_empty_timeline(TIMELINE_ID)?;
-        let writer = tline.writer();
+        let repo = get_test_repo("test_branch")?;
+        let timelineid = ZTimelineId::from_str("11223344556677881122334455667788").unwrap();
+        let tline = repo.create_empty_timeline(timelineid)?;

        // Import initial dummy checkpoint record, otherwise the get_timeline() call
        // after branching fails below
-        writer.put_page_image(RelishTag::Checkpoint, 0, Lsn(0x10), ZERO_CHECKPOINT.clone())?;
+        tline.put_page_image(RelishTag::Checkpoint, 0, Lsn(0x10), ZERO_CHECKPOINT.clone())?;

        // Create a relation on the timeline
-        writer.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
-        writer.put_page_image(TESTREL_A, 0, Lsn(0x30), TEST_IMG("foo blk 0 at 3"))?;
-        writer.put_page_image(TESTREL_A, 0, Lsn(0x40), TEST_IMG("foo blk 0 at 4"))?;
+        tline.put_page_image(TESTREL_A, 0, Lsn(0x20), TEST_IMG("foo blk 0 at 2"))?;
+        tline.put_page_image(TESTREL_A, 0, Lsn(0x30), TEST_IMG("foo blk 0 at 3"))?;
+        tline.put_page_image(TESTREL_A, 0, Lsn(0x40), TEST_IMG("foo blk 0 at 4"))?;

        // Create another relation
-        writer.put_page_image(TESTREL_B, 0, Lsn(0x20), TEST_IMG("foobar blk 0 at 2"))?;
+        tline.put_page_image(TESTREL_B, 0, Lsn(0x20), TEST_IMG("foobar blk 0 at 2"))?;

-        writer.advance_last_record_lsn(Lsn(0x40));
+        tline.advance_last_record_lsn(Lsn(0x40));
        assert_current_logical_size(&tline, Lsn(0x40));

        // Branch the history, modify relation differently on the new timeline
-        repo.branch_timeline(TIMELINE_ID, NEW_TIMELINE_ID, Lsn(0x30))?;
-        let newtline = repo.get_timeline(NEW_TIMELINE_ID)?;
-        let new_writer = newtline.writer();
+        let newtimelineid = ZTimelineId::from_str("AA223344556677881122334455667788").unwrap();
+        repo.branch_timeline(timelineid, newtimelineid, Lsn(0x30))?;
+        let newtline = repo.get_timeline(newtimelineid)?;

-        new_writer.put_page_image(TESTREL_A, 0, Lsn(0x40), TEST_IMG("bar blk 0 at 4"))?;
-        new_writer.advance_last_record_lsn(Lsn(0x40));
+        newtline.put_page_image(TESTREL_A, 0, Lsn(0x40), TEST_IMG("bar blk 0 at 4"))?;
+        newtline.advance_last_record_lsn(Lsn(0x40));

        // Check page contents on both branches
        assert_eq!(
@@ -737,89 +572,8 @@ mod tests {
        Ok(())
    }

-    #[test]
-    fn corrupt_metadata() -> Result<()> {
-        const TEST_NAME: &str = "corrupt_metadata";
-        let harness = RepoHarness::create(TEST_NAME)?;
-        let repo = harness.load();
-
-        repo.create_empty_timeline(TIMELINE_ID)?;
-        drop(repo);
-
-        let metadata_path = harness.timeline_path(&TIMELINE_ID).join(METADATA_FILE_NAME);
-
-        assert!(metadata_path.is_file());
-
-        let mut metadata_bytes = std::fs::read(&metadata_path)?;
-        assert_eq!(metadata_bytes.len(), 512);
-        metadata_bytes[512 - 4 - 2] ^= 1;
-        std::fs::write(metadata_path, metadata_bytes)?;
-
-        let new_repo = harness.load();
-        let err = new_repo.get_timeline(TIMELINE_ID).err().unwrap();
-        assert!(err.to_string().contains("checksum"));
-
-        Ok(())
-    }
-
-    #[test]
-    fn future_layerfiles() -> Result<()> {
-        const TEST_NAME: &str = "future_layerfiles";
-        let harness = RepoHarness::create(TEST_NAME)?;
-        let repo = harness.load();
-
-        repo.create_empty_timeline(TIMELINE_ID)?;
-        drop(repo);
-
-        let timeline_path = harness.timeline_path(&TIMELINE_ID);
-
-        let make_empty_file = |filename: &str| -> std::io::Result<()> {
-            let path = timeline_path.join(filename);
-
-            assert!(!path.exists());
-            std::fs::write(&path, &[])?;
-
-            Ok(())
-        };
-
-        let image_filename = format!("pg_control_0_{:016X}", 8000);
-        let delta_filename = format!("pg_control_0_{:016X}_{:016X}", 8000, 8008);
-
-        make_empty_file(&image_filename)?;
-        make_empty_file(&delta_filename)?;
-
-        let new_repo = harness.load();
-        new_repo.get_timeline(TIMELINE_ID).unwrap();
-        drop(new_repo);
-
-        let check_old = |filename: &str, num: u32| {
-            let path = timeline_path.join(filename);
-            assert!(!path.exists());
-
-            let backup_path = timeline_path.join(format!("{}.{}.old", filename, num));
-            assert!(backup_path.exists());
-        };
-
-        check_old(&image_filename, 0);
-        check_old(&delta_filename, 0);
-
-        make_empty_file(&image_filename)?;
-        make_empty_file(&delta_filename)?;
-
-        let new_repo = harness.load();
-        new_repo.get_timeline(TIMELINE_ID).unwrap();
-        drop(new_repo);
-
-        check_old(&image_filename, 0);
-        check_old(&delta_filename, 0);
-        check_old(&image_filename, 1);
-        check_old(&delta_filename, 1);
-
-        Ok(())
-    }
-
    // Mock WAL redo manager that doesn't do much
-    struct TestRedoManager;
+    struct TestRedoManager {}

    impl WalRedoManager for TestRedoManager {
        fn request_redo(
@@ -828,7 +582,7 @@ mod tests {
            blknum: u32,
            lsn: Lsn,
            base_img: Option<Bytes>,
-            records: Vec<(Lsn, WALRecord)>,
+            records: Vec<WALRecord>,
        ) -> Result<Bytes, WalRedoError> {
            let s = format!(
                "redo for {} blk {} to get to {}, with {} and {} records",
--- a/pageserver/src/restore_local_repo.rs
+++ b/pageserver/src/restore_local_repo.rs
@@ -2,6 +2,7 @@
 //! Import data and WAL from a PostgreSQL data directory and WAL segments into
 //! zenith Timeline.
 //!
+use log::*;
 use postgres_ffi::nonrelfile_utils::clogpage_precedes;
 use postgres_ffi::nonrelfile_utils::slru_may_delete_clogsegment;
 use std::cmp::min;
@@ -10,9 +11,8 @@ use std::fs::File;
 use std::io::Read;
 use std::path::Path;

-use anyhow::{bail, Result};
+use anyhow::Result;
 use bytes::{Buf, Bytes};
-use tracing::*;

 use crate::relish::*;
 use crate::repository::*;
@@ -34,7 +34,7 @@ static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; 8192]);
 ///
 pub fn import_timeline_from_postgres_datadir(
    path: &Path,
-    writer: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    lsn: Lsn,
 ) -> Result<()> {
    // Scan 'global'
@@ -44,10 +44,10 @@ pub fn import_timeline_from_postgres_datadir(
            None => continue,

            Some("pg_control") => {
-                import_control_file(writer, lsn, &direntry.path())?;
+                import_control_file(timeline, lsn, &direntry.path())?;
            }
            Some("pg_filenode.map") => import_nonrel_file(
-                writer,
+                timeline,
                lsn,
                RelishTag::FileNodeMap {
                    spcnode: pg_constants::GLOBALTABLESPACE_OID,
@@ -59,7 +59,7 @@ pub fn import_timeline_from_postgres_datadir(
            // Load any relation files into the page server
            _ => import_relfile(
                &direntry.path(),
-                writer,
+                timeline,
                lsn,
                pg_constants::GLOBALTABLESPACE_OID,
                0,
@@ -86,7 +86,7 @@ pub fn import_timeline_from_postgres_datadir(

                Some("PG_VERSION") => continue,
                Some("pg_filenode.map") => import_nonrel_file(
-                    writer,
+                    timeline,
                    lsn,
                    RelishTag::FileNodeMap {
                        spcnode: pg_constants::DEFAULTTABLESPACE_OID,
@@ -98,7 +98,7 @@ pub fn import_timeline_from_postgres_datadir(
                // Load any relation files into the page server
                _ => import_relfile(
                    &direntry.path(),
-                    writer,
+                    timeline,
                    lsn,
                    pg_constants::DEFAULTTABLESPACE_OID,
                    dboid,
@@ -108,24 +108,24 @@ pub fn import_timeline_from_postgres_datadir(
    }
    for entry in fs::read_dir(path.join("pg_xact"))? {
        let entry = entry?;
-        import_slru_file(writer, lsn, SlruKind::Clog, &entry.path())?;
+        import_slru_file(timeline, lsn, SlruKind::Clog, &entry.path())?;
    }
    for entry in fs::read_dir(path.join("pg_multixact").join("members"))? {
        let entry = entry?;
-        import_slru_file(writer, lsn, SlruKind::MultiXactMembers, &entry.path())?;
+        import_slru_file(timeline, lsn, SlruKind::MultiXactMembers, &entry.path())?;
    }
    for entry in fs::read_dir(path.join("pg_multixact").join("offsets"))? {
        let entry = entry?;
-        import_slru_file(writer, lsn, SlruKind::MultiXactOffsets, &entry.path())?;
+        import_slru_file(timeline, lsn, SlruKind::MultiXactOffsets, &entry.path())?;
    }
    for entry in fs::read_dir(path.join("pg_twophase"))? {
        let entry = entry?;
        let xid = u32::from_str_radix(entry.path().to_str().unwrap(), 16)?;
-        import_nonrel_file(writer, lsn, RelishTag::TwoPhase { xid }, &entry.path())?;
+        import_nonrel_file(timeline, lsn, RelishTag::TwoPhase { xid }, &entry.path())?;
    }
    // TODO: Scan pg_tblspc

-    writer.advance_last_record_lsn(lsn);
+    timeline.advance_last_record_lsn(lsn);

    Ok(())
 }
@@ -133,13 +133,12 @@ pub fn import_timeline_from_postgres_datadir(
 // subroutine of import_timeline_from_postgres_datadir(), to load one relation file.
 fn import_relfile(
    path: &Path,
-    timeline: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    lsn: Lsn,
    spcoid: Oid,
    dboid: Oid,
 ) -> Result<()> {
    // Does it look like a relation file?
-    trace!("importing rel file {}", path.display());

    let p = parse_relfilename(path.file_name().unwrap().to_str().unwrap());
    if let Err(e) = p {
@@ -167,14 +166,15 @@ fn import_relfile(
            }

            // TODO: UnexpectedEof is expected
-            Err(err) => match err.kind() {
+            Err(e) => match e.kind() {
                std::io::ErrorKind::UnexpectedEof => {
                    // reached EOF. That's expected.
                    // FIXME: maybe check that we read the full length of the file?
                    break;
                }
                _ => {
-                    bail!("error reading file {}: {:#}", path.display(), err);
+                    error!("error reading file: {:?} ({})", path, e);
+                    break;
                }
            },
        };
@@ -191,7 +191,7 @@ fn import_relfile(
 /// are just slurped into the repository as one blob.
 ///
 fn import_nonrel_file(
-    timeline: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    lsn: Lsn,
    tag: RelishTag,
    path: &Path,
@@ -201,7 +201,7 @@ fn import_nonrel_file(
    // read the whole file
    file.read_to_end(&mut buffer)?;

-    trace!("importing non-rel file {}", path.display());
+    info!("importing non-rel file {}", path.display());

    timeline.put_page_image(tag, 0, lsn, Bytes::copy_from_slice(&buffer[..]))?;
    Ok(())
@@ -212,13 +212,13 @@ fn import_nonrel_file(
 ///
 /// The control file is imported as is, but we also extract the checkpoint record
 /// from it and store it separated.
-fn import_control_file(timeline: &dyn TimelineWriter, lsn: Lsn, path: &Path) -> Result<()> {
+fn import_control_file(timeline: &dyn Timeline, lsn: Lsn, path: &Path) -> Result<()> {
    let mut file = File::open(path)?;
    let mut buffer = Vec::new();
    // read the whole file
    file.read_to_end(&mut buffer)?;

-    trace!("importing control file {}", path.display());
+    info!("importing control file {}", path.display());

    // Import it as ControlFile
    timeline.put_page_image(
@@ -239,18 +239,13 @@ fn import_control_file(timeline: &dyn TimelineWriter, lsn: Lsn, path: &Path) ->
 ///
 /// Import an SLRU segment file
 ///
-fn import_slru_file(
-    timeline: &dyn TimelineWriter,
-    lsn: Lsn,
-    slru: SlruKind,
-    path: &Path,
-) -> Result<()> {
+fn import_slru_file(timeline: &dyn Timeline, lsn: Lsn, slru: SlruKind, path: &Path) -> Result<()> {
    // Does it look like an SLRU file?
    let mut file = File::open(path)?;
    let mut buf: [u8; 8192] = [0u8; 8192];
    let segno = u32::from_str_radix(path.file_name().unwrap().to_str().unwrap(), 16)?;

-    trace!("importing slru file {}", path.display());
+    info!("importing slru file {}", path.display());

    let mut rpageno = 0;
    loop {
@@ -266,14 +261,15 @@ fn import_slru_file(
            }

            // TODO: UnexpectedEof is expected
-            Err(err) => match err.kind() {
+            Err(e) => match e.kind() {
                std::io::ErrorKind::UnexpectedEof => {
                    // reached EOF. That's expected.
                    // FIXME: maybe check that we read the full length of the file?
                    break;
                }
                _ => {
-                    bail!("error reading file {}: {:#}", path.display(), err);
+                    error!("error reading file: {:?} ({})", path, e);
+                    break;
                }
            },
        };
@@ -291,15 +287,12 @@ fn import_slru_file(
 ///
 pub fn save_decoded_record(
    checkpoint: &mut CheckPoint,
-    checkpoint_modified: &mut bool,
-    timeline: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    decoded: &DecodedWALRecord,
    recdata: Bytes,
    lsn: Lsn,
 ) -> Result<()> {
-    if checkpoint.update_next_xid(decoded.xl_xid) {
-        *checkpoint_modified = true;
-    }
+    checkpoint.update_next_xid(decoded.xl_xid);

    // Iterate through all the blocks that the record modifies, and
    // "put" a separate copy of the record for each block.
@@ -312,12 +305,13 @@ pub fn save_decoded_record(
        });

        let rec = WALRecord {
+            lsn,
            will_init: blk.will_init || blk.apply_image,
            rec: recdata.clone(),
            main_data_offset: decoded.main_data_offset as u32,
        };

-        timeline.put_wal_record(lsn, tag, blk.blkno, rec)?;
+        timeline.put_wal_record(tag, blk.blkno, rec)?;
    }

    let mut buf = decoded.record.clone();
@@ -382,7 +376,7 @@ pub fn save_decoded_record(
        } else {
            assert!(info == pg_constants::CLOG_TRUNCATE);
            let xlrec = XlClogTruncate::decode(&mut buf);
-            save_clog_truncate_record(checkpoint, checkpoint_modified, timeline, lsn, &xlrec)?;
+            save_clog_truncate_record(checkpoint, timeline, lsn, &xlrec)?;
        }
    } else if decoded.xl_rmid == pg_constants::RM_XACT_ID {
        let info = decoded.xl_info & pg_constants::XLOG_XACT_OPMASK;
@@ -451,17 +445,10 @@ pub fn save_decoded_record(
            )?;
        } else if info == pg_constants::XLOG_MULTIXACT_CREATE_ID {
            let xlrec = XlMultiXactCreate::decode(&mut buf);
-            save_multixact_create_record(
-                checkpoint,
-                checkpoint_modified,
-                timeline,
-                lsn,
-                &xlrec,
-                decoded,
-            )?;
+            save_multixact_create_record(checkpoint, timeline, lsn, &xlrec, decoded)?;
        } else if info == pg_constants::XLOG_MULTIXACT_TRUNCATE_ID {
            let xlrec = XlMultiXactTruncate::decode(&mut buf);
-            save_multixact_truncate_record(checkpoint, checkpoint_modified, timeline, lsn, &xlrec)?;
+            save_multixact_truncate_record(checkpoint, timeline, lsn, &xlrec)?;
        }
    } else if decoded.xl_rmid == pg_constants::RM_RELMAP_ID {
        let xlrec = XlRelmapUpdate::decode(&mut buf);
@@ -470,10 +457,7 @@ pub fn save_decoded_record(
        let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
        if info == pg_constants::XLOG_NEXTOID {
            let next_oid = buf.get_u32_le();
-            if checkpoint.nextOid != next_oid {
-                checkpoint.nextOid = next_oid;
-                *checkpoint_modified = true;
-            }
+            checkpoint.nextOid = next_oid;
        } else if info == pg_constants::XLOG_CHECKPOINT_ONLINE
            || info == pg_constants::XLOG_CHECKPOINT_SHUTDOWN
        {
@@ -489,7 +473,6 @@ pub fn save_decoded_record(
            );
            if (checkpoint.oldestXid.wrapping_sub(xlog_checkpoint.oldestXid) as i32) < 0 {
                checkpoint.oldestXid = xlog_checkpoint.oldestXid;
-                *checkpoint_modified = true;
            }
        }
    }
@@ -497,11 +480,7 @@ pub fn save_decoded_record(
 }

 /// Subroutine of save_decoded_record(), to handle an XLOG_DBASE_CREATE record.
-fn save_xlog_dbase_create(
-    timeline: &dyn TimelineWriter,
-    lsn: Lsn,
-    rec: &XlCreateDatabase,
-) -> Result<()> {
+fn save_xlog_dbase_create(timeline: &dyn Timeline, lsn: Lsn, rec: &XlCreateDatabase) -> Result<()> {
    let db_id = rec.db_id;
    let tablespace_id = rec.tablespace_id;
    let src_db_id = rec.src_db_id;
@@ -578,11 +557,7 @@ fn save_xlog_dbase_create(
 /// Subroutine of save_decoded_record(), to handle an XLOG_SMGR_TRUNCATE record.
 ///
 /// This is the same logic as in PostgreSQL's smgr_redo() function.
-fn save_xlog_smgr_truncate(
-    timeline: &dyn TimelineWriter,
-    lsn: Lsn,
-    rec: &XlSmgrTruncate,
-) -> Result<()> {
+fn save_xlog_smgr_truncate(timeline: &dyn Timeline, lsn: Lsn, rec: &XlSmgrTruncate) -> Result<()> {
    let spcnode = rec.rnode.spcnode;
    let dbnode = rec.rnode.dbnode;
    let relnode = rec.rnode.relnode;
@@ -644,7 +619,7 @@ fn save_xlog_smgr_truncate(
 /// Subroutine of save_decoded_record(), to handle an XLOG_XACT_* records.
 ///
 fn save_xact_record(
-    timeline: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    lsn: Lsn,
    parsed: &XlXactParsedRecord,
    decoded: &DecodedWALRecord,
@@ -655,12 +630,12 @@ fn save_xact_record(
    let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
    let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
    let rec = WALRecord {
+        lsn,
        will_init: false,
        rec: decoded.record.clone(),
        main_data_offset: decoded.main_data_offset as u32,
    };
    timeline.put_wal_record(
-        lsn,
        RelishTag::Slru {
            slru: SlruKind::Clog,
            segno,
@@ -676,7 +651,6 @@ fn save_xact_record(
            let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
            let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
            timeline.put_wal_record(
-                lsn,
                RelishTag::Slru {
                    slru: SlruKind::Clog,
                    segno,
@@ -702,8 +676,7 @@ fn save_xact_record(

 fn save_clog_truncate_record(
    checkpoint: &mut CheckPoint,
-    checkpoint_modified: &mut bool,
-    timeline: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    lsn: Lsn,
    xlrec: &XlClogTruncate,
 ) -> Result<()> {
@@ -721,7 +694,6 @@ fn save_clog_truncate_record(
    // TODO Figure out if there will be any issues with replica.
    checkpoint.oldestXid = xlrec.oldest_xid;
    checkpoint.oldestXidDB = xlrec.oldest_xid_db;
-    *checkpoint_modified = true;

    // TODO Treat AdvanceOldestClogXid() or write a comment why we don't need it

@@ -764,13 +736,13 @@ fn save_clog_truncate_record(

 fn save_multixact_create_record(
    checkpoint: &mut CheckPoint,
-    checkpoint_modified: &mut bool,
-    timeline: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    lsn: Lsn,
    xlrec: &XlMultiXactCreate,
    decoded: &DecodedWALRecord,
 ) -> Result<()> {
    let rec = WALRecord {
+        lsn,
        will_init: false,
        rec: decoded.record.clone(),
        main_data_offset: decoded.main_data_offset as u32,
@@ -779,7 +751,6 @@ fn save_multixact_create_record(
    let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
    let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
    timeline.put_wal_record(
-        lsn,
        RelishTag::Slru {
            slru: SlruKind::MultiXactOffsets,
            segno,
@@ -799,7 +770,6 @@ fn save_multixact_create_record(
        let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
        let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
        timeline.put_wal_record(
-            lsn,
            RelishTag::Slru {
                slru: SlruKind::MultiXactMembers,
                segno,
@@ -822,11 +792,9 @@ fn save_multixact_create_record(
    }
    if xlrec.mid >= checkpoint.nextMulti {
        checkpoint.nextMulti = xlrec.mid + 1;
-        *checkpoint_modified = true;
    }
    if xlrec.moff + xlrec.nmembers > checkpoint.nextMultiOffset {
        checkpoint.nextMultiOffset = xlrec.moff + xlrec.nmembers;
-        *checkpoint_modified = true;
    }
    let max_mbr_xid = xlrec.members.iter().fold(0u32, |acc, mbr| {
        if mbr.xid.wrapping_sub(acc) as i32 > 0 {
@@ -836,22 +804,18 @@ fn save_multixact_create_record(
        }
    });

-    if checkpoint.update_next_xid(max_mbr_xid) {
-        *checkpoint_modified = true;
-    }
+    checkpoint.update_next_xid(max_mbr_xid);
    Ok(())
 }

 fn save_multixact_truncate_record(
    checkpoint: &mut CheckPoint,
-    checkpoint_modified: &mut bool,
-    timeline: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    lsn: Lsn,
    xlrec: &XlMultiXactTruncate,
 ) -> Result<()> {
    checkpoint.oldestMulti = xlrec.end_trunc_off;
    checkpoint.oldestMultiDB = xlrec.oldest_multi_db;
-    *checkpoint_modified = true;

    // PerformMembersTruncation
    let maxsegment: i32 = mx_offset_to_member_segment(pg_constants::MAX_MULTIXACT_OFFSET);
@@ -885,7 +849,7 @@ fn save_multixact_truncate_record(
 }

 fn save_relmap_page(
-    timeline: &dyn TimelineWriter,
+    timeline: &dyn Timeline,
    lsn: Lsn,
    xlrec: &XlRelmapUpdate,
    decoded: &DecodedWALRecord,
--- a/pageserver/src/tenant_mgr.rs
+++ b/pageserver/src/tenant_mgr.rs
@@ -8,139 +8,47 @@ use crate::walredo::PostgresRedoManager;
 use crate::PageServerConf;
 use anyhow::{anyhow, bail, Context, Result};
 use lazy_static::lazy_static;
-use log::{debug, info};
-use std::collections::hash_map::Entry;
+use log::info;
 use std::collections::HashMap;
 use std::fs;
 use std::str::FromStr;
-use std::sync::atomic::{AtomicBool, Ordering};
-use std::sync::{Arc, Mutex, MutexGuard};
-use std::thread::JoinHandle;
+use std::sync::{Arc, Mutex};
 use zenith_utils::zid::{ZTenantId, ZTimelineId};

 lazy_static! {
-    static ref REPOSITORY: Mutex<HashMap<ZTenantId, Arc<dyn Repository>>> =
+    pub static ref REPOSITORY: Mutex<HashMap<ZTenantId, Arc<dyn Repository>>> =
        Mutex::new(HashMap::new());
 }

-fn access_repository() -> MutexGuard<'static, HashMap<ZTenantId, Arc<dyn Repository>>> {
-    REPOSITORY.lock().unwrap()
-}
-struct TenantHandleEntry {
-    checkpointer_handle: Option<JoinHandle<()>>,
-    gc_handle: Option<JoinHandle<()>>,
-}
-
-// Logically these handles belong to Repository,
-// but it's just simpler to store them separately
-lazy_static! {
-    static ref TENANT_HANDLES: Mutex<HashMap<ZTenantId, TenantHandleEntry>> =
-        Mutex::new(HashMap::new());
-}
-
-static SHUTDOWN_REQUESTED: AtomicBool = AtomicBool::new(false);
-
 pub fn init(conf: &'static PageServerConf) {
-    let mut m = access_repository();
+    let mut m = REPOSITORY.lock().unwrap();
+
    for dir_entry in fs::read_dir(conf.tenants_path()).unwrap() {
        let tenantid =
            ZTenantId::from_str(dir_entry.unwrap().file_name().to_str().unwrap()).unwrap();
-        let repo = init_repo(conf, tenantid);
+
+        // Set up a WAL redo manager, for applying WAL records.
+        let walredo_mgr = PostgresRedoManager::new(conf, tenantid);
+
+        // Set up an object repository, for actual data storage.
+        let repo = Arc::new(LayeredRepository::new(
+            conf,
+            Arc::new(walredo_mgr),
+            tenantid,
+        ));
+        LayeredRepository::launch_checkpointer_thread(conf, repo.clone());
+        LayeredRepository::launch_gc_thread(conf, repo.clone());
+
        info!("initialized storage for tenant: {}", &tenantid);
        m.insert(tenantid, repo);
    }
 }

-fn init_repo(conf: &'static PageServerConf, tenant_id: ZTenantId) -> Arc<LayeredRepository> {
-    // Set up a WAL redo manager, for applying WAL records.
-    let walredo_mgr = PostgresRedoManager::new(conf, tenant_id);
-
-    // Set up an object repository, for actual data storage.
-    let repo = Arc::new(LayeredRepository::new(
-        conf,
-        Arc::new(walredo_mgr),
-        tenant_id,
-        true,
-    ));
-
-    let checkpointer_handle = LayeredRepository::launch_checkpointer_thread(conf, repo.clone());
-    let gc_handle = LayeredRepository::launch_gc_thread(conf, repo.clone());
-
-    let mut handles = TENANT_HANDLES.lock().unwrap();
-    let h = TenantHandleEntry {
-        checkpointer_handle: Some(checkpointer_handle),
-        gc_handle: Some(gc_handle),
-    };
-
-    handles.insert(tenant_id, h);
-
-    repo
-}
-
-// TODO kb Currently unused function, will later be used when the relish storage downloads a new layer.
-// Relevant PR: https://github.com/zenithdb/zenith/pull/686
-pub fn register_relish_download(
-    conf: &'static PageServerConf,
-    tenant_id: ZTenantId,
-    timeline_id: ZTimelineId,
-) {
-    log::info!(
-        "Registering new download, tenant id {}, timeline id: {}",
-        tenant_id,
-        timeline_id
-    );
-    match access_repository().entry(tenant_id) {
-        Entry::Occupied(o) => init_timeline(o.get().as_ref(), timeline_id),
-        Entry::Vacant(v) => {
-            log::info!("New repo initialized");
-            let new_repo = init_repo(conf, tenant_id);
-            init_timeline(new_repo.as_ref(), timeline_id);
-            v.insert(new_repo);
-        }
-    }
-}
-
-fn init_timeline(repo: &dyn Repository, timeline_id: ZTimelineId) {
-    match repo.get_timeline(timeline_id) {
-        Ok(_timeline) => log::info!("Successfully initialized timeline {}", timeline_id),
-        Err(e) => log::error!("Failed to init timeline {}, reason: {:#}", timeline_id, e),
-    }
-}
-
-// Check this flag in the thread loops to know when to exit
-pub fn shutdown_requested() -> bool {
-    SHUTDOWN_REQUESTED.load(Ordering::Relaxed)
-}
-
-pub fn stop_tenant_threads(tenantid: ZTenantId) {
-    let mut handles = TENANT_HANDLES.lock().unwrap();
-    if let Some(h) = handles.get_mut(&tenantid) {
-        h.checkpointer_handle.take().map(JoinHandle::join);
-        debug!("checkpointer for tenant {} has stopped", tenantid);
-        h.gc_handle.take().map(JoinHandle::join);
-        debug!("gc for tenant {} has stopped", tenantid);
-    }
-}
-
-pub fn shutdown_all_tenants() -> Result<()> {
-    SHUTDOWN_REQUESTED.swap(true, Ordering::Relaxed);
-
-    let tenants = list_tenants()?;
-    for tenantid in tenants {
-        stop_tenant_threads(tenantid);
-        let repo = get_repository_for_tenant(tenantid)?;
-        debug!("shutdown tenant {}", tenantid);
-        repo.shutdown()?;
-    }
-
-    Ok(())
-}
-
 pub fn create_repository_for_tenant(
    conf: &'static PageServerConf,
    tenantid: ZTenantId,
 ) -> Result<()> {
-    let mut m = access_repository();
+    let mut m = REPOSITORY.lock().unwrap();

    // First check that the tenant doesn't exist already
    if m.get(&tenantid).is_some() {
@@ -154,10 +62,15 @@ pub fn create_repository_for_tenant(
    Ok(())
 }

+pub fn insert_repository_for_tenant(tenantid: ZTenantId, repo: Arc<dyn Repository>) {
+    let o = &mut REPOSITORY.lock().unwrap();
+    o.insert(tenantid, repo);
+}
+
 pub fn get_repository_for_tenant(tenantid: ZTenantId) -> Result<Arc<dyn Repository>> {
-    access_repository()
-        .get(&tenantid)
-        .map(Arc::clone)
+    let o = &REPOSITORY.lock().unwrap();
+    o.get(&tenantid)
+        .map(|repo| Arc::clone(repo))
        .ok_or_else(|| anyhow!("repository not found for tenant name {}", tenantid))
 }

@@ -169,14 +82,3 @@ pub fn get_timeline_for_tenant(
        .get_timeline(timelineid)
        .with_context(|| format!("cannot fetch timeline {}", timelineid))
 }
-
-fn list_tenants() -> Result<Vec<ZTenantId>> {
-    let o = &mut REPOSITORY.lock().unwrap();
-
-    o.iter()
-        .map(|tenant| {
-            let (tenantid, _) = tenant;
-            Ok(*tenantid)
-        })
-        .collect()
-}
--- a/pageserver/src/waldecoder.rs
+++ b/pageserver/src/waldecoder.rs
@@ -72,10 +72,6 @@ impl WalStreamDecoder {
    ///     Err(WalDecodeError): an error occured while decoding, meaning the input was invalid.
    ///
    pub fn poll_decode(&mut self) -> Result<Option<(Lsn, Bytes)>, WalDecodeError> {
-        let recordbuf;
-
-        // Run state machine that validates page headers, and reassembles records
-        // that cross page boundaries.
        loop {
            // parse and verify page boundaries as we go
            if self.lsn.segment_offset(pg_constants::WAL_SEGMENT_SIZE) == 0 {
@@ -124,41 +120,29 @@ impl WalStreamDecoder {
                self.lsn += self.padlen as u64;
                self.padlen = 0;
            } else if self.contlen == 0 {
-                assert!(self.recordbuf.is_empty());
-
                // need to have at least the xl_tot_len field
+
                if self.inputbuf.remaining() < 4 {
                    return Ok(None);
                }

-                // peek xl_tot_len at the beginning of the record.
-                // FIXME: assumes little-endian
+                // read xl_tot_len FIXME: assumes little-endian
                self.startlsn = self.lsn;
-                let xl_tot_len = (&self.inputbuf[0..4]).get_u32_le();
+                let xl_tot_len = self.inputbuf.get_u32_le();
                if (xl_tot_len as usize) < XLOG_SIZE_OF_XLOG_RECORD {
                    return Err(WalDecodeError {
                        msg: format!("invalid xl_tot_len {}", xl_tot_len),
                        lsn: self.lsn,
                    });
                }
+                self.lsn += 4;

-                // Fast path for the common case that the whole record fits on the page.
-                let pageleft = self.lsn.remaining_in_block() as u32;
-                if self.inputbuf.remaining() >= xl_tot_len as usize && xl_tot_len <= pageleft {
-                    // Take the record from the 'inputbuf', and validate it.
-                    recordbuf = self.inputbuf.copy_to_bytes(xl_tot_len as usize);
-                    self.lsn += xl_tot_len as u64;
-                    break;
-                } else {
-                    // Need to assemble the record from pieces. Remember the size of the
-                    // record, and loop back. On next iteration, we will reach the 'else'
-                    // branch below, and copy the part of the record that was on this page
-                    // to 'recordbuf'.  Subsequent iterations will skip page headers, and
-                    // append the continuations from the next pages to 'recordbuf'.
-                    self.recordbuf.reserve(xl_tot_len as usize);
-                    self.contlen = xl_tot_len;
-                    continue;
-                }
+                self.recordbuf.clear();
+                self.recordbuf.reserve(xl_tot_len as usize);
+                self.recordbuf.put_u32_le(xl_tot_len);
+
+                self.contlen = xl_tot_len - 4;
+                continue;
            } else {
                // we're continuing a record, possibly from previous page.
                let pageleft = self.lsn.remaining_in_block() as u32;
@@ -175,42 +159,47 @@ impl WalStreamDecoder {
                self.contlen -= n as u32;

                if self.contlen == 0 {
-                    // The record is now complete.
-                    recordbuf = std::mem::replace(&mut self.recordbuf, BytesMut::new()).freeze();
-                    break;
+                    let recordbuf = std::mem::replace(&mut self.recordbuf, BytesMut::new());
+
+                    let recordbuf = recordbuf.freeze();
+                    let mut buf = recordbuf.clone();
+
+                    let xlogrec = XLogRecord::from_bytes(&mut buf);
+
+                    // XLOG_SWITCH records are special. If we see one, we need to skip
+                    // to the next WAL segment.
+                    if xlogrec.is_xlog_switch_record() {
+                        trace!("saw xlog switch record at {}", self.lsn);
+                        self.padlen =
+                            self.lsn.calc_padding(pg_constants::WAL_SEGMENT_SIZE as u64) as u32;
+                    } else {
+                        // Pad to an 8-byte boundary
+                        self.padlen = self.lsn.calc_padding(8u32) as u32;
+                    }
+
+                    let mut crc = crc32c_append(0, &recordbuf[XLOG_RECORD_CRC_OFFS + 4..]);
+                    crc = crc32c_append(crc, &recordbuf[0..XLOG_RECORD_CRC_OFFS]);
+                    if crc != xlogrec.xl_crc {
+                        return Err(WalDecodeError {
+                            msg: "WAL record crc mismatch".into(),
+                            lsn: self.lsn,
+                        });
+                    }
+
+                    // Always align resulting LSN on 0x8 boundary -- that is important for getPage()
+                    // and WalReceiver integration. Since this code is used both for WalReceiver and
+                    // initial WAL import let's force alignment right here.
+                    let result = (self.lsn.align(), recordbuf);
+                    return Ok(Some(result));
                }
                continue;
            }
        }
+        // check record boundaries

-        // We now have a record in the 'recordbuf' local variable.
-        let xlogrec = XLogRecord::from_slice(&recordbuf[0..XLOG_SIZE_OF_XLOG_RECORD]);
+        // deal with continuation records

-        let mut crc = 0;
-        crc = crc32c_append(crc, &recordbuf[XLOG_RECORD_CRC_OFFS + 4..]);
-        crc = crc32c_append(crc, &recordbuf[0..XLOG_RECORD_CRC_OFFS]);
-        if crc != xlogrec.xl_crc {
-            return Err(WalDecodeError {
-                msg: "WAL record crc mismatch".into(),
-                lsn: self.lsn,
-            });
-        }
-
-        // XLOG_SWITCH records are special. If we see one, we need to skip
-        // to the next WAL segment.
-        if xlogrec.is_xlog_switch_record() {
-            trace!("saw xlog switch record at {}", self.lsn);
-            self.padlen = self.lsn.calc_padding(pg_constants::WAL_SEGMENT_SIZE as u64) as u32;
-        } else {
-            // Pad to an 8-byte boundary
-            self.padlen = self.lsn.calc_padding(8u32) as u32;
-        }
-
-        // Always align resulting LSN on 0x8 boundary -- that is important for getPage()
-        // and WalReceiver integration. Since this code is used both for WalReceiver and
-        // initial WAL import let's force alignment right here.
-        let result = (self.lsn.align(), recordbuf);
-        Ok(Some(result))
+        // deal with xlog_switch records
    }
 }

--- a/pageserver/src/walreceiver.rs
+++ b/pageserver/src/walreceiver.rs
@@ -10,24 +10,24 @@ use crate::restore_local_repo;
 use crate::tenant_mgr;
 use crate::waldecoder::*;
 use crate::PageServerConf;
-use anyhow::{bail, Error, Result};
+use anyhow::{Error, Result};
 use lazy_static::lazy_static;
+use log::*;
 use postgres::fallible_iterator::FallibleIterator;
 use postgres::replication::ReplicationIter;
 use postgres::{Client, NoTls, SimpleQueryMessage, SimpleQueryRow};
+use postgres_ffi::xlog_utils::*;
 use postgres_ffi::*;
 use postgres_protocol::message::backend::ReplicationMessage;
 use postgres_types::PgLsn;
-use std::cell::Cell;
+use std::cmp::{max, min};
 use std::collections::HashMap;
+use std::fs;
 use std::str::FromStr;
 use std::sync::Mutex;
 use std::thread;
 use std::thread::sleep;
-use std::thread::JoinHandle;
-use std::thread_local;
 use std::time::{Duration, SystemTime};
-use tracing::*;
 use zenith_utils::lsn::Lsn;
 use zenith_utils::zid::ZTenantId;
 use zenith_utils::zid::ZTimelineId;
@@ -37,7 +37,6 @@ use zenith_utils::zid::ZTimelineId;
 //
 struct WalReceiverEntry {
    wal_producer_connstr: String,
-    wal_receiver_handle: Option<JoinHandle<()>>,
 }

 lazy_static! {
@@ -45,26 +44,6 @@ lazy_static! {
        Mutex::new(HashMap::new());
 }

-thread_local! {
-    // Boolean that is true only for WAL receiver threads
-    //
-    // This is used in `wait_lsn` to guard against usage that might lead to a deadlock.
-    pub(crate) static IS_WAL_RECEIVER: Cell<bool> = Cell::new(false);
-}
-
-// Wait for walreceiver to stop
-// Now it stops when pageserver shutdown is requested.
-// In future we can make this more granular and send shutdown signals
-// per tenant/timeline to cancel inactive walreceivers.
-// TODO deal with blocking pg connections
-pub fn stop_wal_receiver(timelineid: ZTimelineId) {
-    let mut receivers = WAL_RECEIVERS.lock().unwrap();
-    if let Some(r) = receivers.get_mut(&timelineid) {
-        r.wal_receiver_handle.take();
-        // r.wal_receiver_handle.take().map(JoinHandle::join);
-    }
-}
-
 // Launch a new WAL receiver, or tell one that's running about change in connection string
 pub fn launch_wal_receiver(
    conf: &'static PageServerConf,
@@ -79,19 +58,21 @@ pub fn launch_wal_receiver(
            receiver.wal_producer_connstr = wal_producer_connstr.into();
        }
        None => {
-            let wal_receiver_handle = thread::Builder::new()
+            let receiver = WalReceiverEntry {
+                wal_producer_connstr: wal_producer_connstr.into(),
+            };
+            receivers.insert(timelineid, receiver);
+
+            // Also launch a new thread to handle this connection
+            //
+            // NOTE: This thread name is checked in the assertion in wait_lsn. If you change
+            // this, make sure you update the assertion too.
+            let _walreceiver_thread = thread::Builder::new()
                .name("WAL receiver thread".into())
                .spawn(move || {
-                    IS_WAL_RECEIVER.with(|c| c.set(true));
                    thread_main(conf, timelineid, tenantid);
                })
                .unwrap();
-
-            let receiver = WalReceiverEntry {
-                wal_producer_connstr: wal_producer_connstr.into(),
-                wal_receiver_handle: Some(wal_receiver_handle),
-            };
-            receivers.insert(timelineid, receiver);
        }
    };
 }
@@ -111,14 +92,16 @@ fn get_wal_producer_connstr(timelineid: ZTimelineId) -> String {
 // This is the entry point for the WAL receiver thread.
 //
 fn thread_main(conf: &'static PageServerConf, timelineid: ZTimelineId, tenantid: ZTenantId) {
-    let _enter = info_span!("WAL receiver", timeline = %timelineid, tenant = %tenantid).entered();
-    info!("WAL receiver thread started");
+    info!(
+        "WAL receiver thread started for timeline : '{}'",
+        timelineid
+    );

    //
    // Make a connection to the WAL safekeeper, or directly to the primary PostgreSQL server,
    // and start streaming WAL from it. If the connection is lost, keep retrying.
    //
-    while !tenant_mgr::shutdown_requested() {
+    loop {
        // Look up the current WAL producer address
        let wal_producer_connstr = get_wal_producer_connstr(timelineid);

@@ -132,11 +115,10 @@ fn thread_main(conf: &'static PageServerConf, timelineid: ZTimelineId, tenantid:
            sleep(Duration::from_secs(1));
        }
    }
-    debug!("WAL streaming shut down");
 }

 fn walreceiver_main(
-    _conf: &PageServerConf,
+    conf: &PageServerConf,
    timelineid: ZTimelineId,
    wal_producer_connstr: &str,
    tenantid: ZTenantId,
@@ -176,15 +158,15 @@ fn walreceiver_main(
    let mut startpoint = last_rec_lsn;

    if startpoint == Lsn(0) {
-        bail!("No previous WAL position");
+        error!("No previous WAL position");
    }

    // There might be some padding after the last full record, skip it.
    startpoint += startpoint.calc_padding(8u32);

    info!(
-        "last_record_lsn {} starting replication from {}, server is at {}...",
-        last_rec_lsn, startpoint, end_of_wal
+        "last_record_lsn {} starting replication from {} for timeline {}, server is at {}...",
+        last_rec_lsn, startpoint, timelineid, end_of_wal
    );

    let query = format!("START_REPLICATION PHYSICAL {}", startpoint);
@@ -206,38 +188,34 @@ fn walreceiver_main(
                let data = xlog_data.data();
                let startlsn = Lsn::from(xlog_data.wal_start());
                let endlsn = startlsn + data.len() as u64;
+                let prev_last_rec_lsn = last_rec_lsn;

                trace!("received XLogData between {} and {}", startlsn, endlsn);

                waldecoder.feed_bytes(data);

                while let Some((lsn, recdata)) = waldecoder.poll_decode()? {
-                    let _enter = info_span!("processing record", lsn = %lsn).entered();
+                    // Save old checkpoint value to compare with it after decoding WAL record
+                    let old_checkpoint_bytes = checkpoint.encode();
+                    let decoded = decode_wal_record(recdata.clone());

                    // It is important to deal with the aligned records as lsn in getPage@LSN is
                    // aligned and can be several bytes bigger. Without this alignment we are
                    // at risk of hittind a deadlock.
                    assert!(lsn.is_aligned());

-                    let writer = timeline.writer();
-
-                    let mut checkpoint_modified = false;
-
-                    let decoded = decode_wal_record(recdata.clone());
                    restore_local_repo::save_decoded_record(
                        &mut checkpoint,
-                        &mut checkpoint_modified,
-                        writer.as_ref(),
+                        &*timeline,
                        &decoded,
                        recdata,
                        lsn,
                    )?;

+                    let new_checkpoint_bytes = checkpoint.encode();
                    // Check if checkpoint data was updated by save_decoded_record
-                    if checkpoint_modified {
-                        let new_checkpoint_bytes = checkpoint.encode();
-
-                        writer.put_page_image(
+                    if new_checkpoint_bytes != old_checkpoint_bytes {
+                        timeline.put_page_image(
                            RelishTag::Checkpoint,
                            0,
                            lsn,
@@ -247,10 +225,38 @@ fn walreceiver_main(

                    // Now that this record has been fully handled, including updating the
                    // checkpoint data, let the repository know that it is up-to-date to this LSN
-                    writer.advance_last_record_lsn(lsn);
+                    timeline.advance_last_record_lsn(lsn);
                    last_rec_lsn = lsn;
                }

+                // Somewhat arbitrarily, if we have at least 10 complete wal segments (16 MB each),
+                // "checkpoint" the repository to flush all the changes from WAL we've processed
+                // so far to disk. After this, we don't need the original WAL anymore, and it
+                // can be removed. This is probably too aggressive for production, but it's useful
+                // to expose bugs now.
+                //
+                // TODO: We don't actually dare to remove the WAL. It's useful for debugging,
+                // and we might it for logical decoding other things in the future. Although
+                // we should also be able to fetch it back from the WAL safekeepers or S3 if
+                // needed.
+                if prev_last_rec_lsn.segment_number(pg_constants::WAL_SEGMENT_SIZE)
+                    != last_rec_lsn.segment_number(pg_constants::WAL_SEGMENT_SIZE)
+                {
+                    info!("switched segment {} to {}", prev_last_rec_lsn, last_rec_lsn);
+                    let (oldest_segno, newest_segno) = find_wal_file_range(
+                        conf,
+                        &timelineid,
+                        pg_constants::WAL_SEGMENT_SIZE,
+                        last_rec_lsn,
+                        &tenantid,
+                    )?;
+
+                    if newest_segno - oldest_segno >= 10 {
+                        // TODO: This is where we could remove WAL older than last_rec_lsn.
+                        //remove_wal_files(timelineid, pg_constants::WAL_SEGMENT_SIZE, last_rec_lsn)?;
+                    }
+                }
+
                if !caught_up && endlsn >= end_of_wal {
                    info!("caught up at LSN {}", endlsn);
                    caught_up = true;
@@ -272,7 +278,7 @@ fn walreceiver_main(
                );

                if reply_requested {
-                    Some(last_rec_lsn)
+                    Some(timeline.get_last_record_lsn())
                } else {
                    None
                }
@@ -292,15 +298,51 @@ fn walreceiver_main(

            physical_stream.standby_status_update(write_lsn, flush_lsn, apply_lsn, ts, NO_REPLY)?;
        }
-
-        if tenant_mgr::shutdown_requested() {
-            debug!("stop walreceiver because pageserver shutdown is requested");
-            break;
-        }
    }
    Ok(())
 }

+fn find_wal_file_range(
+    conf: &PageServerConf,
+    timeline: &ZTimelineId,
+    wal_seg_size: usize,
+    written_upto: Lsn,
+    tenant: &ZTenantId,
+) -> Result<(u64, u64)> {
+    let written_upto_segno = written_upto.segment_number(wal_seg_size);
+
+    let mut oldest_segno = written_upto_segno;
+    let mut newest_segno = written_upto_segno;
+    // Scan the wal directory, and count how many WAL filed we could remove
+    let wal_dir = conf.wal_dir_path(timeline, tenant);
+    for entry in fs::read_dir(wal_dir)? {
+        let entry = entry?;
+        let path = entry.path();
+
+        if path.is_dir() {
+            continue;
+        }
+
+        let filename = path.file_name().unwrap().to_str().unwrap();
+
+        if IsXLogFileName(filename) {
+            let (segno, _tli) = XLogFromFileName(filename, wal_seg_size);
+
+            if segno > written_upto_segno {
+                // that's strange.
+                warn!("there is a WAL file from future at {}", path.display());
+                continue;
+            }
+
+            oldest_segno = min(oldest_segno, segno);
+            newest_segno = max(newest_segno, segno);
+        }
+    }
+    // FIXME: would be good to assert that there are no gaps in the WAL files
+
+    Ok((oldest_segno, newest_segno))
+}
+
 /// Data returned from the postgres `IDENTIFY_SYSTEM` command
 ///
 /// See the [postgres docs] for more details.
--- a/pageserver/src/walredo.rs
+++ b/pageserver/src/walredo.rs
@@ -22,7 +22,8 @@ use byteorder::{ByteOrder, LittleEndian};
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 use lazy_static::lazy_static;
 use log::*;
-use serde::Serialize;
+use serde::{Deserialize, Serialize};
+use std::cell::RefCell;
 use std::fs;
 use std::fs::OpenOptions;
 use std::io::prelude::*;
@@ -59,7 +60,7 @@ use postgres_ffi::XLogRecord;
 /// In Postgres `BufferTag` structure is used for exactly the same purpose.
 /// [See more related comments here](https://github.com/postgres/postgres/blob/99c5852e20a0987eca1c38ba0c09329d4076b6a0/src/include/storage/buf_internals.h#L91).
 ///
-#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone, Copy, Serialize)]
+#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone, Copy, Serialize, Deserialize)]
 pub struct BufferTag {
    pub rel: RelTag,
    pub blknum: u32,
@@ -82,7 +83,7 @@ pub trait WalRedoManager: Send + Sync {
        blknum: u32,
        lsn: Lsn,
        base_img: Option<Bytes>,
-        records: Vec<(Lsn, WALRecord)>,
+        records: Vec<WALRecord>,
    ) -> Result<Bytes, WalRedoError>;
 }

@@ -99,7 +100,7 @@ impl crate::walredo::WalRedoManager for DummyRedoManager {
        _blknum: u32,
        _lsn: Lsn,
        _base_img: Option<Bytes>,
-        _records: Vec<(Lsn, WALRecord)>,
+        _records: Vec<WALRecord>,
    ) -> Result<Bytes, WalRedoError> {
        Err(WalRedoError::InvalidState)
    }
@@ -150,7 +151,7 @@ struct WalRedoRequest {
    lsn: Lsn,

    base_img: Option<Bytes>,
-    records: Vec<(Lsn, WALRecord)>,
+    records: Vec<WALRecord>,
 }

 /// An error happened in WAL redo
@@ -179,7 +180,7 @@ impl WalRedoManager for PostgresRedoManager {
        blknum: u32,
        lsn: Lsn,
        base_img: Option<Bytes>,
-        records: Vec<(Lsn, WALRecord)>,
+        records: Vec<WALRecord>,
    ) -> Result<Bytes, WalRedoError> {
        let start_time;
        let lock_time;
@@ -205,7 +206,7 @@ impl WalRedoManager for PostgresRedoManager {
                    .block_on(PostgresRedoProcess::launch(self.conf, &self.tenantid))?;
                *process_guard = Some(p);
            }
-            let process = process_guard.as_mut().unwrap();
+            let process = (*process_guard).as_ref().unwrap();

            self.runtime
                .block_on(self.handle_apply_request(process, &request))
@@ -246,7 +247,7 @@ impl PostgresRedoManager {
    ///
    async fn handle_apply_request(
        &self,
-        process: &mut PostgresRedoProcess,
+        process: &PostgresRedoProcess,
        request: &WalRedoRequest,
    ) -> Result<Bytes, WalRedoError> {
        let rel = request.rel;
@@ -277,7 +278,7 @@ impl PostgresRedoManager {
                page.extend_from_slice(&ZERO_PAGE);
            }
            // Apply all collected WAL records
-            for (_lsn, record) in records {
+            for record in records {
                let mut buf = record.rec.clone();

                WAL_REDO_RECORD_COUNTER.inc();
@@ -298,11 +299,9 @@ impl PostgresRedoManager {
                    // Transaction manager stuff
                    let rec_segno = match rel {
                        RelishTag::Slru { slru, segno } => {
-                            assert!(
-                                slru == SlruKind::Clog,
-                                "Not valid XACT relish tag {:?}",
-                                rel
-                            );
+                            if slru != SlruKind::Clog {
+                                panic!("Not valid XACT relish tag {:?}", rel);
+                            }
                            segno
                        }
                        _ => panic!("Not valid XACT relish tag {:?}", rel),
@@ -422,7 +421,7 @@ impl PostgresRedoManager {
        );

        if let Err(e) = apply_result {
-            error!("could not apply WAL records: {:#}", e);
+            error!("could not apply WAL records: {}", e);
            result = Err(WalRedoError::IoError(e));
        } else {
            let img = apply_result.unwrap();
@@ -439,8 +438,8 @@ impl PostgresRedoManager {
 /// Handle to the Postgres WAL redo process
 ///
 struct PostgresRedoProcess {
-    stdin: ChildStdin,
-    stdout: ChildStdout,
+    stdin: RefCell<ChildStdin>,
+    stdout: RefCell<ChildStdout>,
 }

 impl PostgresRedoProcess {
@@ -460,7 +459,7 @@ impl PostgresRedoProcess {
        if datadir.exists() {
            info!("directory {:?} exists, removing", &datadir);
            if let Err(e) = fs::remove_dir_all(&datadir) {
-                error!("could not remove old wal-redo-datadir: {:#}", e);
+                error!("could not remove old wal-redo-datadir: {:?}", e);
            }
        }
        info!("running initdb in {:?}", datadir.display());
@@ -533,7 +532,10 @@ impl PostgresRedoProcess {
        };
        tokio::spawn(f_stderr);

-        Ok(PostgresRedoProcess { stdin, stdout })
+        Ok(PostgresRedoProcess {
+            stdin: RefCell::new(stdin),
+            stdout: RefCell::new(stdout),
+        })
    }

    //
@@ -541,14 +543,13 @@ impl PostgresRedoProcess {
    // new page image.
    //
    async fn apply_wal_records(
-        &mut self,
+        &self,
        tag: BufferTag,
        base_img: Option<Bytes>,
-        records: &[(Lsn, WALRecord)],
+        records: &[WALRecord],
    ) -> Result<Bytes, std::io::Error> {
-        let stdout = &mut self.stdout;
-        // Buffer the writes to avoid a lot of small syscalls.
-        let mut stdin = tokio::io::BufWriter::new(&mut self.stdin);
+        let mut stdin = self.stdin.borrow_mut();
+        let mut stdout = self.stdout.borrow_mut();

        // We do three things simultaneously: send the old base image and WAL records to
        // the child process's stdin, read the result from child's stdout, and forward any logging
@@ -565,16 +566,22 @@ impl PostgresRedoProcess {
                stdin.write_all(&build_begin_redo_for_block_msg(tag)),
            )
            .await??;
-            if let Some(img) = base_img {
-                timeout(TIMEOUT, stdin.write_all(&build_push_page_msg(tag, &img))).await??;
+            if base_img.is_some() {
+                timeout(
+                    TIMEOUT,
+                    stdin.write_all(&build_push_page_msg(tag, base_img.unwrap())),
+                )
+                .await??;
            }

            // Send WAL records.
-            for (lsn, rec) in records.iter() {
+            for rec in records.iter() {
+                let r = rec.clone();
+
                WAL_REDO_RECORD_COUNTER.inc();

                stdin
-                    .write_all(&build_apply_record_msg(*lsn, &rec.rec))
+                    .write_all(&build_apply_record_msg(r.lsn, r.rec))
                    .await?;

                //debug!("sent WAL record to wal redo postgres process ({:X}/{:X}",
@@ -611,41 +618,58 @@ impl PostgresRedoProcess {
 // process. See vendor/postgres/src/backend/tcop/zenith_wal_redo.c for
 // explanation of the protocol.

-fn build_begin_redo_for_block_msg(tag: BufferTag) -> Vec<u8> {
+fn build_begin_redo_for_block_msg(tag: BufferTag) -> Bytes {
    let len = 4 + 1 + 4 * 4;
-    let mut buf = Vec::with_capacity(1 + len);
+    let mut buf = BytesMut::with_capacity(1 + len);

    buf.put_u8(b'B');
    buf.put_u32(len as u32);

-    tag.ser_into(&mut buf)
+    // FIXME: this is a temporary hack that should go away when we refactor
+    // the postgres protocol serialization + handlers.
+    //
+    // BytesMut is a dynamic growable buffer, used a lot in tokio code but
+    // not in the std library. To write to a BytesMut from a serde serializer,
+    // we need to either:
+    // - pre-allocate the required buffer space. This is annoying because we
+    //   shouldn't care what the exact serialized size is-- that's the
+    //   serializer's job.
+    // - Or, we need to create a temporary "writer" (which implements the
+    //   `Write` trait). It's a bit awkward, because the writer consumes the
+    //   underlying BytesMut, and we need to extract it later with
+    //   `into_inner`.
+    let mut writer = buf.writer();
+    tag.ser_into(&mut writer)
        .expect("serialize BufferTag should always succeed");
+    let buf = writer.into_inner();

    debug_assert!(buf.len() == 1 + len);

-    buf
+    buf.freeze()
 }

-fn build_push_page_msg(tag: BufferTag, base_img: &[u8]) -> Vec<u8> {
+fn build_push_page_msg(tag: BufferTag, base_img: Bytes) -> Bytes {
    assert!(base_img.len() == 8192);

    let len = 4 + 1 + 4 * 4 + base_img.len();
-    let mut buf = Vec::with_capacity(1 + len);
+    let mut buf = BytesMut::with_capacity(1 + len);

    buf.put_u8(b'P');
    buf.put_u32(len as u32);
-    tag.ser_into(&mut buf)
+    let mut writer = buf.writer();
+    tag.ser_into(&mut writer)
        .expect("serialize BufferTag should always succeed");
+    let mut buf = writer.into_inner();
    buf.put(base_img);

    debug_assert!(buf.len() == 1 + len);

-    buf
+    buf.freeze()
 }

-fn build_apply_record_msg(endlsn: Lsn, rec: &[u8]) -> Vec<u8> {
+fn build_apply_record_msg(endlsn: Lsn, rec: Bytes) -> Bytes {
    let len = 4 + 8 + rec.len();
-    let mut buf: Vec<u8> = Vec::with_capacity(1 + len);
+    let mut buf = BytesMut::with_capacity(1 + len);

    buf.put_u8(b'A');
    buf.put_u32(len as u32);
@@ -654,19 +678,21 @@ fn build_apply_record_msg(endlsn: Lsn, rec: &[u8]) -> Vec<u8> {

    debug_assert!(buf.len() == 1 + len);

-    buf
+    buf.freeze()
 }

-fn build_get_page_msg(tag: BufferTag) -> Vec<u8> {
+fn build_get_page_msg(tag: BufferTag) -> Bytes {
    let len = 4 + 1 + 4 * 4;
-    let mut buf = Vec::with_capacity(1 + len);
+    let mut buf = BytesMut::with_capacity(1 + len);

    buf.put_u8(b'G');
    buf.put_u32(len as u32);
-    tag.ser_into(&mut buf)
+    let mut writer = buf.writer();
+    tag.ser_into(&mut writer)
        .expect("serialize BufferTag should always succeed");
+    let buf = writer.into_inner();

    debug_assert!(buf.len() == 1 + len);

-    buf
+    buf.freeze()
 }
--- a/postgres_ffi/src/xlog_utils.rs
+++ b/postgres_ffi/src/xlog_utils.rs
@@ -329,12 +329,7 @@ pub fn main() {
 }

 impl XLogRecord {
-    pub fn from_slice(buf: &[u8]) -> XLogRecord {
-        use zenith_utils::bin_ser::LeSer;
-        XLogRecord::des(buf).unwrap()
-    }
-
-    pub fn from_bytes<B: Buf>(buf: &mut B) -> XLogRecord {
+    pub fn from_bytes(buf: &mut Bytes) -> XLogRecord {
        use zenith_utils::bin_ser::LeSer;
        XLogRecord::des_from(&mut buf.reader()).unwrap()
    }
@@ -382,12 +377,10 @@ impl CheckPoint {
        Ok(CheckPoint::des(buf)?)
    }

-    /// Update next XID based on provided new_xid and stored epoch.
-    /// Next XID should be greater than new_xid. This handles 32-bit
-    /// XID wraparound correctly.
-    ///
-    /// Returns 'true' if the XID was updated.
-    pub fn update_next_xid(&mut self, xid: u32) -> bool {
+    // Update next XID based on provided new_xid and stored epoch.
+    // Next XID should be greater than new_xid.
+    // Also take in account 32-bit wrap-around.
+    pub fn update_next_xid(&mut self, xid: u32) {
        let xid = xid.wrapping_add(XID_CHECKPOINT_INTERVAL - 1) & !(XID_CHECKPOINT_INTERVAL - 1);
        let full_xid = self.nextXid.value;
        let new_xid = std::cmp::max(xid + 1, pg_constants::FIRST_NORMAL_TRANSACTION_ID);
@@ -398,14 +391,10 @@ impl CheckPoint {
                // wrap-around
                epoch += 1;
            }
-            let nextXid = (epoch << 32) | new_xid as u64;
-
-            if nextXid != self.nextXid.value {
-                self.nextXid = FullTransactionId { value: nextXid };
-                return true;
-            }
+            self.nextXid = FullTransactionId {
+                value: (epoch << 32) | new_xid as u64,
+            };
        }
-        false
    }
 }

--- a/proxy/Cargo.toml
+++ b/proxy/Cargo.toml
@@ -18,6 +18,5 @@ tokio = "1.11"
 tokio-postgres = { git = "https://github.com/zenithdb/rust-postgres.git", rev="9eb0dbfbeb6a6c1b79099b9f7ae4a8c021877858" }
 clap = "2.33.0"
 rustls = "0.19.1"
-reqwest = { version = "0.11", features = ["blocking", "json"] }

 zenith_utils = { path = "../zenith_utils" }
--- a/proxy/src/cplane_api.rs
+++ b/proxy/src/cplane_api.rs
@@ -1,14 +1,17 @@
-use anyhow::{bail, Context, Result};
+use anyhow::{bail, Result};
 use serde::{Deserialize, Serialize};
-use std::net::{SocketAddr, ToSocketAddrs};
+use std::{
+    collections::HashMap,
+    net::{IpAddr, SocketAddr},
+};

 pub struct CPlaneApi {
-    auth_endpoint: &'static str,
+    // address: SocketAddr,
 }

-#[derive(Serialize, Deserialize, Debug)]
+#[derive(Serialize, Deserialize)]
 pub struct DatabaseInfo {
-    pub host: String,
+    pub host: IpAddr, // TODO: allow host name here too
    pub port: u16,
    pub dbname: String,
    pub user: String,
@@ -16,13 +19,8 @@ pub struct DatabaseInfo {
 }

 impl DatabaseInfo {
-    pub fn socket_addr(&self) -> Result<SocketAddr> {
-        let host_port = format!("{}:{}", self.host, self.port);
-        host_port
-            .to_socket_addrs()
-            .with_context(|| format!("cannot resolve {} to SocketAddr", host_port))?
-            .next()
-            .ok_or_else(|| anyhow::Error::msg("cannot resolve at least one SocketAddr"))
+    pub fn socket_addr(&self) -> SocketAddr {
+        SocketAddr::new(self.host, self.port)
    }

    pub fn conn_string(&self) -> String {
@@ -33,35 +31,62 @@ impl DatabaseInfo {
    }
 }

+// mock cplane api
 impl CPlaneApi {
-    pub fn new(auth_endpoint: &'static str) -> CPlaneApi {
-        CPlaneApi { auth_endpoint }
+    pub fn new(_address: &SocketAddr) -> CPlaneApi {
+        CPlaneApi {
+            // address: address.clone(),
+        }
    }

-    pub fn authenticate_proxy_request(
-        &self,
-        user: &str,
-        database: &str,
-        md5_response: &[u8],
-        salt: &[u8; 4],
-    ) -> Result<DatabaseInfo> {
-        let mut url = reqwest::Url::parse(self.auth_endpoint)?;
-        url.query_pairs_mut()
-            .append_pair("login", user)
-            .append_pair("database", database)
-            .append_pair("md5response", std::str::from_utf8(md5_response)?)
-            .append_pair("salt", &hex::encode(salt));
+    pub fn check_auth(&self, user: &str, md5_response: &[u8], salt: &[u8; 4]) -> Result<()> {
+        // passwords for both is "mypass"
+        let auth_map: HashMap<_, &str> = vec![
+            ("stas@zenith", "716ee6e1c4a9364d66285452c47402b1"),
+            ("stas2@zenith", "3996f75df64c16a8bfaf01301b61d582"),
+        ]
+        .into_iter()
+        .collect();

-        println!("cplane request: {}", url.as_str());
+        let stored_hash = auth_map
+            .get(&user)
+            .ok_or_else(|| anyhow::Error::msg("user not found"))?;
+        let salted_stored_hash = format!(
+            "md5{:x}",
+            md5::compute([stored_hash.as_bytes(), salt].concat())
+        );

-        let resp = reqwest::blocking::get(url)?;
+        let received_hash = std::str::from_utf8(md5_response)?;

-        if resp.status().is_success() {
-            let conn_info: DatabaseInfo = serde_json::from_str(resp.text()?.as_str())?;
-            println!("got conn info: #{:?}", conn_info);
-            Ok(conn_info)
+        println!(
+            "auth: {} rh={} sh={} ssh={} {:?}",
+            user, received_hash, stored_hash, salted_stored_hash, salt
+        );
+
+        if received_hash == salted_stored_hash {
+            Ok(())
        } else {
            bail!("Auth failed")
        }
    }
+
+    pub fn get_database_uri(&self, _user: &str, _database: &str) -> Result<DatabaseInfo> {
+        Ok(DatabaseInfo {
+            host: "127.0.0.1".parse()?,
+            port: 5432,
+            dbname: "stas".to_string(),
+            user: "stas".to_string(),
+            password: "mypass".to_string(),
+        })
+    }
+
+    // pub fn create_database(&self, _user: &String, _database: &String) -> Result<DatabaseInfo> {
+    //     Ok(DatabaseInfo {
+    //         host: "127.0.0.1".parse()?,
+    //         port: 5432,
+    //         dbname: "stas".to_string(),
+    //         user: "stas".to_string(),
+    //         password: "mypass".to_string(),
+    //     })
+    // }
 }
--- a/proxy/src/main.rs
+++ b/proxy/src/main.rs
@@ -34,7 +34,7 @@ pub struct ProxyConf {
    pub redirect_uri: String,

    /// control plane address where we would check auth.
-    pub auth_endpoint: String,
+    pub cplane_address: SocketAddr,

    pub ssl_config: Option<Arc<ServerConfig>>,
 }
@@ -56,7 +56,8 @@ fn configure_ssl(arg_matches: &ArgMatches) -> anyhow::Result<Option<Arc<ServerCo

    let key = {
        let key_bytes = std::fs::read(key_path).context("SSL key file")?;
-        let mut keys = pemfile::pkcs8_private_keys(&mut &key_bytes[..])
+        let mut keys = pemfile::rsa_private_keys(&mut &key_bytes[..])
+            .or_else(|_| pemfile::pkcs8_private_keys(&mut &key_bytes[..]))
            .map_err(|_| anyhow!("couldn't read TLS keys"))?;
        ensure!(keys.len() == 1, "keys.len() = {} (should be 1)", keys.len());
        keys.pop().unwrap()
@@ -99,15 +100,7 @@ fn main() -> anyhow::Result<()> {
                .long("uri")
                .takes_value(true)
                .help("redirect unauthenticated users to given uri")
-                .default_value("http://127.0.0.1:3000/psql_session/"),
-        )
-        .arg(
-            Arg::with_name("auth-endpoint")
-                .short("a")
-                .long("auth-endpoint")
-                .takes_value(true)
-                .help("redirect unauthenticated users to given uri")
-                .default_value("http://127.0.0.1:3000/authenticate_proxy_request/"),
+                .default_value("http://localhost:3000/psql_session/"),
        )
        .arg(
            Arg::with_name("ssl-key")
@@ -129,7 +122,7 @@ fn main() -> anyhow::Result<()> {
        proxy_address: arg_matches.value_of("proxy").unwrap().parse()?,
        mgmt_address: arg_matches.value_of("mgmt").unwrap().parse()?,
        redirect_uri: arg_matches.value_of("uri").unwrap().parse()?,
-        auth_endpoint: arg_matches.value_of("auth-endpoint").unwrap().parse()?,
+        cplane_address: "127.0.0.1:3000".parse()?,
        ssl_config: configure_ssl(&arg_matches)?,
    };
    let state = ProxyState {
--- a/proxy/src/mgmt.rs
+++ b/proxy/src/mgmt.rs
@@ -5,7 +5,7 @@ use std::{

 use anyhow::bail;
 use bytes::Bytes;
-use serde::Deserialize;
+use serde::{Deserialize, Serialize};
 use zenith_utils::{
    postgres_backend::{self, query_from_cstring, AuthType, PostgresBackend},
    pq_proto::{BeMessage, SINGLE_COL_ROWDESC},
@@ -34,7 +34,7 @@ pub fn thread_main(state: &'static ProxyState, listener: TcpListener) -> anyhow:

 pub fn mgmt_conn_main(state: &'static ProxyState, socket: TcpStream) -> anyhow::Result<()> {
    let mut conn_handler = MgmtHandler { state };
-    let pgbackend = PostgresBackend::new(socket, AuthType::Trust, None, true)?;
+    let pgbackend = PostgresBackend::new(socket, AuthType::Trust, None)?;
    pgbackend.run(&mut conn_handler)
 }

@@ -49,7 +49,7 @@ struct MgmtHandler {
 //             "host": "127.0.0.1",
 //             "port": 5432,
 //             "dbname": "stas",
-//             "user": "stas",
+//             "user": "stas"
 //             "password": "mypass"
 //         }
 //     }
@@ -60,16 +60,13 @@ struct MgmtHandler {
 //         "Failure": "oops"
 //     }
 // }
-//
-// // to test manually by sending a query to mgmt interface:
-// psql -h 127.0.0.1 -p 9999 -c '{"session_id":"4f10dde522e14739","result":{"Success":{"host":"127.0.0.1","port":5432,"dbname":"stas","user":"stas","password":"stas"}}}'
-#[derive(Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct PsqlSessionResponse {
    session_id: String,
    result: PsqlSessionResult,
 }

-#[derive(Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum PsqlSessionResult {
    Success(DatabaseInfo),
    Failure(String),
@@ -81,47 +78,34 @@ impl postgres_backend::Handler for MgmtHandler {
        pgb: &mut PostgresBackend,
        query_string: Bytes,
    ) -> anyhow::Result<()> {
-        let res = try_process_query(self, pgb, query_string);
-        // intercept and log error message
-        if res.is_err() {
-            println!("Mgmt query failed: #{:?}", res);
-        }
-        res
-    }
-}
+        let query_string = query_from_cstring(query_string);

-fn try_process_query(
-    mgmt: &mut MgmtHandler,
-    pgb: &mut PostgresBackend,
-    query_string: Bytes,
-) -> anyhow::Result<()> {
-    let query_string = query_from_cstring(query_string);
+        println!("Got mgmt query: '{}'", std::str::from_utf8(&query_string)?);

-    println!("Got mgmt query: '{}'", std::str::from_utf8(&query_string)?);
+        let resp: PsqlSessionResponse = serde_json::from_slice(&query_string)?;

-    let resp: PsqlSessionResponse = serde_json::from_slice(&query_string)?;
+        let waiters = self.state.waiters.lock().unwrap();

-    let waiters = mgmt.state.waiters.lock().unwrap();
+        let sender = waiters
+            .get(&resp.session_id)
+            .ok_or_else(|| anyhow::Error::msg("psql_session_id is not found"))?;

-    let sender = waiters
-        .get(&resp.session_id)
-        .ok_or_else(|| anyhow::Error::msg("psql_session_id is not found"))?;
+        match resp.result {
+            PsqlSessionResult::Success(db_info) => {
+                sender.send(Ok(db_info))?;

-    match resp.result {
-        PsqlSessionResult::Success(db_info) => {
-            sender.send(Ok(db_info))?;
+                pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?
+                    .write_message_noflush(&BeMessage::DataRow(&[Some(b"ok")]))?
+                    .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
+                pgb.flush()?;
+                Ok(())
+            }

-            pgb.write_message_noflush(&SINGLE_COL_ROWDESC)?
-                .write_message_noflush(&BeMessage::DataRow(&[Some(b"ok")]))?
-                .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
-            pgb.flush()?;
-            Ok(())
-        }
+            PsqlSessionResult::Failure(message) => {
+                sender.send(Err(anyhow::Error::msg(message.clone())))?;

-        PsqlSessionResult::Failure(message) => {
-            sender.send(Err(anyhow::Error::msg(message.clone())))?;
-
-            bail!("psql session request failed: {}", message)
+                bail!("psql session request failed: {}", message)
+            }
        }
    }
 }
--- a/proxy/src/proxy.rs
+++ b/proxy/src/proxy.rs
@@ -57,14 +57,13 @@ pub fn proxy_conn_main(
 ) -> anyhow::Result<()> {
    let mut conn = ProxyConnection {
        state,
-        cplane: CPlaneApi::new(&state.conf.auth_endpoint),
+        cplane: CPlaneApi::new(&state.conf.cplane_address),
        user: "".into(),
        database: "".into(),
        pgb: PostgresBackend::new(
            socket,
            postgres_backend::AuthType::MD5,
            state.conf.ssl_config.clone(),
-            false,
        )?,
        md5_salt: [0u8; 4],
        psql_session_id: "".into(),
@@ -81,8 +80,6 @@ pub fn proxy_conn_main(
        conn.handle_new_user()?
    };

-    // XXX: move that inside handle_new_user/handle_existing_user to be able to
-    // report wrong connection error.
    proxy_pass(conn.pgb, db_info)
 }

@@ -175,31 +172,21 @@ impl ProxyConnection {
                .split_last()
                .ok_or_else(|| anyhow::Error::msg("unexpected password message"))?;

-            match self.cplane.authenticate_proxy_request(
-                self.user.as_str(),
-                self.database.as_str(),
-                md5_response,
-                &self.md5_salt,
-            ) {
-                Err(e) => {
-                    self.pgb
-                        .write_message(&BeMessage::ErrorResponse(format!("{}", e)))?;
-
-                    bail!("auth failed: {}", e);
-                }
-                Ok(conn_info) => {
-                    self.pgb
-                        .write_message_noflush(&BeMessage::AuthenticationOk)?;
-                    self.pgb
-                        .write_message_noflush(&BeMessage::ParameterStatus)?;
-                    self.pgb.write_message(&BeMessage::ReadyForQuery)?;
-
-                    Ok(conn_info)
-                }
+            if let Err(e) = self.check_auth_md5(md5_response) {
+                self.pgb
+                    .write_message(&BeMessage::ErrorResponse(format!("{}", e)))?;
+                bail!("auth failed: {}", e);
+            } else {
+                self.pgb
+                    .write_message_noflush(&BeMessage::AuthenticationOk)?;
+                self.pgb
+                    .write_message_noflush(&BeMessage::ParameterStatus)?;
+                self.pgb.write_message(&BeMessage::ReadyForQuery)?;
            }
-        } else {
-            bail!("protocol violation");
        }
+
+        // ok, we are authorized
+        self.cplane.get_database_uri(&self.user, &self.database)
    }

    fn handle_new_user(&mut self) -> anyhow::Result<DatabaseInfo> {
@@ -245,11 +232,17 @@ databases without opening the browser.

        Ok(dbinfo)
    }
+
+    fn check_auth_md5(&self, md5_response: &[u8]) -> anyhow::Result<()> {
+        assert!(self.is_existing_user());
+        self.cplane
+            .check_auth(self.user.as_str(), md5_response, &self.md5_salt)
+    }
 }

 /// Create a TCP connection to a postgres database, authenticate with it, and receive the ReadyForQuery message
 async fn connect_to_db(db_info: DatabaseInfo) -> anyhow::Result<tokio::net::TcpStream> {
-    let mut socket = tokio::net::TcpStream::connect(db_info.socket_addr()?).await?;
+    let mut socket = tokio::net::TcpStream::connect(db_info.socket_addr()).await?;
    let config = db_info.conn_string().parse::<tokio_postgres::Config>()?;
    let _ = config.connect_raw(&mut socket, NoTls).await?;
    Ok(socket)
@@ -280,9 +273,7 @@ fn proxy(

 /// Proxy a client connection to a postgres database
 fn proxy_pass(pgb: PostgresBackend, db_info: DatabaseInfo) -> anyhow::Result<()> {
-    let runtime = tokio::runtime::Builder::new_current_thread()
-        .enable_all()
-        .build()?;
+    let runtime = tokio::runtime::Builder::new_current_thread().build()?;
    let db_stream = runtime.block_on(connect_to_db(db_info))?;
    let db_stream = db_stream.into_std()?;
    db_stream.set_nonblocking(false)?;
--- a/test_runner/batch_others/test_branch_behind.py
+++ b/test_runner/batch_others/test_branch_behind.py
@@ -1,6 +1,6 @@
 import subprocess
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log
+

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -13,7 +13,7 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
    zenith_cli.run(["branch", "test_branch_behind", "empty"])

    pgmain = postgres.create_start('test_branch_behind')
-    log.info("postgres is running on 'test_branch_behind' branch")
+    print("postgres is running on 'test_branch_behind' branch")

    main_pg_conn = pgmain.connect()
    main_cur = main_pg_conn.cursor()
@@ -27,17 +27,17 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
    ''')
    main_cur.execute('SELECT pg_current_wal_insert_lsn()')
    lsn_a = main_cur.fetchone()[0]
-    log.info(f'LSN after 100 rows: {lsn_a}')
+    print('LSN after 100 rows: ' + lsn_a)

    # Insert some more rows. (This generates enough WAL to fill a few segments.)
    main_cur.execute('''
        INSERT INTO foo
            SELECT 'long string to consume some space' || g
-            FROM generate_series(1, 200000) g
+            FROM generate_series(1, 100000) g
    ''')
    main_cur.execute('SELECT pg_current_wal_insert_lsn()')
    lsn_b = main_cur.fetchone()[0]
-    log.info(f'LSN after 200100 rows: {lsn_b}')
+    print('LSN after 100100 rows: ' + lsn_b)

    # Branch at the point where only 100 rows were inserted
    zenith_cli.run(["branch", "test_branch_behind_hundred", "test_branch_behind@" + lsn_a])
@@ -46,15 +46,15 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
    main_cur.execute('''
        INSERT INTO foo
            SELECT 'long string to consume some space' || g
-            FROM generate_series(1, 200000) g
+            FROM generate_series(1, 100000) g
    ''')
    main_cur.execute('SELECT pg_current_wal_insert_lsn()')

    main_cur.execute('SELECT pg_current_wal_insert_lsn()')
    lsn_c = main_cur.fetchone()[0]
-    log.info(f'LSN after 400100 rows: {lsn_c}')
+    print('LSN after 200100 rows: ' + lsn_c)

-    # Branch at the point where only 200100 rows were inserted
+    # Branch at the point where only 200 rows were inserted
    zenith_cli.run(["branch", "test_branch_behind_more", "test_branch_behind@" + lsn_b])

    pg_hundred = postgres.create_start("test_branch_behind_hundred")
@@ -70,11 +70,11 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
    more_pg_conn = pg_more.connect()
    more_cur = more_pg_conn.cursor()
    more_cur.execute('SELECT count(*) FROM foo')
-    assert more_cur.fetchone() == (200100, )
+    assert more_cur.fetchone() == (100100, )

    # All the rows are visible on the main branch
    main_cur.execute('SELECT count(*) FROM foo')
-    assert main_cur.fetchone() == (400100, )
+    assert main_cur.fetchone() == (200100, )

    # Check bad lsn's for branching

@@ -89,4 +89,4 @@ def test_branch_behind(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
    try:
        zenith_cli.run(["branch", "test_branch_preinitdb", "test_branch_behind@0/42"])
    except subprocess.CalledProcessError:
-        log.info("Branch creation with pre-initdb LSN failed (as expected)")
+        print("Branch creation with pre-initdb LSN failed (as expected)")
--- a/test_runner/batch_others/test_clog_truncate.py
+++ b/test_runner/batch_others/test_clog_truncate.py
@@ -4,7 +4,6 @@ import os
 from contextlib import closing

 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -25,7 +24,7 @@ def test_clog_truncate(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
    ]

    pg = postgres.create_start('test_clog_truncate', config_lines=config)
-    log.info('postgres is running on test_clog_truncate branch')
+    print('postgres is running on test_clog_truncate branch')

    # Install extension containing function needed for test
    pg.safe_psql('CREATE EXTENSION zenith_test_utils')
@@ -34,22 +33,22 @@ def test_clog_truncate(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
    with closing(pg.connect()) as conn:
        with conn.cursor() as cur:
            cur.execute('select test_consume_xids(1000*1000*10);')
-            log.info('xids consumed')
+            print('xids consumed')

            # call a checkpoint to trigger TruncateSubtrans
            cur.execute('CHECKPOINT;')

            # ensure WAL flush
            cur.execute('select txid_current()')
-            log.info(cur.fetchone())
+            print(cur.fetchone())

    # wait for autovacuum to truncate the pg_xact
    # XXX Is it worth to add a timeout here?
    pg_xact_0000_path = os.path.join(pg.pg_xact_dir_path(), '0000')
-    log.info(f"pg_xact_0000_path = {pg_xact_0000_path}")
+    print("pg_xact_0000_path = " + pg_xact_0000_path)

    while os.path.isfile(pg_xact_0000_path):
-        log.info(f"file exists. wait for truncation. " "pg_xact_0000_path = {pg_xact_0000_path}")
+        print("file exists. wait for truncation. " "pg_xact_0000_path = " + pg_xact_0000_path)
        time.sleep(5)

    # checkpoint to advance latest lsn
@@ -60,14 +59,14 @@ def test_clog_truncate(zenith_cli, pageserver: ZenithPageserver, postgres: Postg
            lsn_after_truncation = cur.fetchone()[0]

    # create new branch after clog truncation and start a compute node on it
-    log.info(f'create branch at lsn_after_truncation {lsn_after_truncation}')
+    print('create branch at lsn_after_truncation ' + lsn_after_truncation)
    zenith_cli.run(
        ["branch", "test_clog_truncate_new", "test_clog_truncate@" + lsn_after_truncation])

    pg2 = postgres.create_start('test_clog_truncate_new')
-    log.info('postgres is running on test_clog_truncate_new branch')
+    print('postgres is running on test_clog_truncate_new branch')

    # check that new node doesn't contain truncated segment
    pg_xact_0000_path_new = os.path.join(pg2.pg_xact_dir_path(), '0000')
-    log.info(f"pg_xact_0000_path_new = {pg_xact_0000_path_new}")
+    print("pg_xact_0000_path_new = " + pg_xact_0000_path_new)
    assert os.path.isfile(pg_xact_0000_path_new) is False
--- a/test_runner/batch_others/test_config.py
+++ b/test_runner/batch_others/test_config.py
@@ -1,7 +1,6 @@
 from contextlib import closing

 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -15,7 +14,7 @@ def test_config(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFact

    # change config
    pg = postgres.create_start('test_config', config_lines=['log_min_messages=debug1'])
-    log.info('postgres is running on test_config branch')
+    print('postgres is running on test_config branch')

    with closing(pg.connect()) as conn:
        with conn.cursor() as cur:
--- a/test_runner/batch_others/test_createdropdb.py
+++ b/test_runner/batch_others/test_createdropdb.py
@@ -3,7 +3,6 @@ import pathlib

 from contextlib import closing
 from fixtures.zenith_fixtures import ZenithPageserver, PostgresFactory, ZenithCli, check_restored_datadir_content
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -20,7 +19,7 @@ def test_createdb(
    zenith_cli.run(["branch", "test_createdb", "empty"])

    pg = postgres.create_start('test_createdb')
-    log.info("postgres is running on 'test_createdb' branch")
+    print("postgres is running on 'test_createdb' branch")

    with closing(pg.connect()) as conn:
        with conn.cursor() as cur:
@@ -54,7 +53,7 @@ def test_dropdb(
    zenith_cli.run(["branch", "test_dropdb", "empty"])

    pg = postgres.create_start('test_dropdb')
-    log.info("postgres is running on 'test_dropdb' branch")
+    print("postgres is running on 'test_dropdb' branch")

    with closing(pg.connect()) as conn:
        with conn.cursor() as cur:
@@ -89,13 +88,13 @@ def test_dropdb(

    # Test that database subdir exists on the branch before drop
    dbpath = pathlib.Path(pg_before.pgdata_dir) / 'base' / str(dboid)
-    log.info(dbpath)
+    print(dbpath)

    assert os.path.isdir(dbpath) == True

    # Test that database subdir doesn't exist on the branch after drop
    dbpath = pathlib.Path(pg_after.pgdata_dir) / 'base' / str(dboid)
-    log.info(dbpath)
+    print(dbpath)

    assert os.path.isdir(dbpath) == False

--- a/test_runner/batch_others/test_createuser.py
+++ b/test_runner/batch_others/test_createuser.py
@@ -1,7 +1,6 @@
 from contextlib import closing

 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -13,7 +12,7 @@ def test_createuser(zenith_cli, pageserver: ZenithPageserver, postgres: Postgres
    zenith_cli.run(["branch", "test_createuser", "empty"])

    pg = postgres.create_start('test_createuser')
-    log.info("postgres is running on 'test_createuser' branch")
+    print("postgres is running on 'test_createuser' branch")

    with closing(pg.connect()) as conn:
        with conn.cursor() as cur:
--- a/test_runner/batch_others/test_multixact.py
+++ b/test_runner/batch_others/test_multixact.py
@@ -1,5 +1,4 @@
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver, check_restored_datadir_content
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -16,7 +15,7 @@ def test_multixact(pageserver: ZenithPageserver, postgres: PostgresFactory,
    zenith_cli.run(["branch", "test_multixact", "empty"])
    pg = postgres.create_start('test_multixact')

-    log.info("postgres is running on 'test_multixact' branch")
+    print("postgres is running on 'test_multixact' branch")
    pg_conn = pg.connect()
    cur = pg_conn.cursor()

@@ -56,7 +55,7 @@ def test_multixact(pageserver: ZenithPageserver, postgres: PostgresFactory,
    zenith_cli.run(["branch", "test_multixact_new", "test_multixact@" + lsn])
    pg_new = postgres.create_start('test_multixact_new')

-    log.info("postgres is running on 'test_multixact_new' branch")
+    print("postgres is running on 'test_multixact_new' branch")
    pg_new_conn = pg_new.connect()
    cur_new = pg_new_conn.cursor()

--- a/test_runner/batch_others/test_old_request_lsn.py
+++ b/test_runner/batch_others/test_old_request_lsn.py
@@ -1,7 +1,6 @@
 from contextlib import closing

 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -19,7 +18,7 @@ def test_old_request_lsn(zenith_cli, pageserver: ZenithPageserver, postgres: Pos
    # Create a branch for us
    zenith_cli.run(["branch", "test_old_request_lsn", "empty"])
    pg = postgres.create_start('test_old_request_lsn')
-    log.info('postgres is running on test_old_request_lsn branch')
+    print('postgres is running on test_old_request_lsn branch')

    pg_conn = pg.connect()
    cur = pg_conn.cursor()
@@ -47,7 +46,7 @@ def test_old_request_lsn(zenith_cli, pageserver: ZenithPageserver, postgres: Pos
        from pg_settings where name = 'shared_buffers'
    ''')
    row = cur.fetchone()
-    log.info(f'shared_buffers is {row[0]}, table size {row[1]}');
+    print("shared_buffers is {}, table size {}", row[0], row[1]);
    assert int(row[0]) < int(row[1])

    cur.execute('VACUUM foo');
--- a/test_runner/batch_others/test_pageserver_restart.py
+++ b/test_runner/batch_others/test_pageserver_restart.py
@@ -5,7 +5,6 @@ import time
 from contextlib import closing
 from multiprocessing import Process, Value
 from fixtures.zenith_fixtures import WalAcceptorFactory, ZenithPageserver, PostgresFactory
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -41,7 +40,7 @@ def test_pageserver_restart(zenith_cli, pageserver: ZenithPageserver, postgres:
        from pg_settings where name = 'shared_buffers'
    ''')
    row = cur.fetchone()
-    log.info(f"shared_buffers is {row[0]}, table size {row[1]}");
+    print("shared_buffers is {}, table size {}", row[0], row[1]);
    assert int(row[0]) < int(row[1])

    # Stop and restart pageserver. This is a more or less graceful shutdown, although
--- a/test_runner/batch_others/test_pgbench.py
+++ b/test_runner/batch_others/test_pgbench.py
@@ -1,5 +1,4 @@
 from fixtures.zenith_fixtures import PostgresFactory
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -9,9 +8,9 @@ def test_pgbench(postgres: PostgresFactory, pg_bin, zenith_cli):
    zenith_cli.run(["branch", "test_pgbench", "empty"])

    pg = postgres.create_start('test_pgbench')
-    log.info("postgres is running on 'test_pgbench' branch")
+    print("postgres is running on 'test_pgbench' branch")

    connstr = pg.connstr()

-    pg_bin.run_capture(['pgbench', '-i', connstr])
-    pg_bin.run_capture(['pgbench'] + '-c 10 -T 5 -P 1 -M prepared'.split() + [connstr])
+    pg_bin.run_capture(['pgbench', '-i', '-s', '100', connstr])
+    pg_bin.run_capture(['pgbench'] + '-c 1 -N -T 100 -P 1 -M prepared'.split() + [connstr])
--- a/test_runner/batch_others/test_restart_compute.py
+++ b/test_runner/batch_others/test_restart_compute.py
@@ -2,7 +2,6 @@ import pytest

 from contextlib import closing
 from fixtures.zenith_fixtures import ZenithPageserver, PostgresFactory
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -10,7 +9,10 @@ pytest_plugins = ("fixtures.zenith_fixtures")
 #
 # Test restarting and recreating a postgres instance
 #
-@pytest.mark.parametrize('with_wal_acceptors', [False, True])
+# XXX: with_wal_acceptors=True fails now, would be fixed with
+# `postgres --sync-walkeepers` patches.
+#
+@pytest.mark.parametrize('with_wal_acceptors', [False])
 def test_restart_compute(
        zenith_cli,
        pageserver: ZenithPageserver,
@@ -28,7 +30,7 @@ def test_restart_compute(

    pg = postgres.create_start('test_restart_compute',
                               wal_acceptors=wal_acceptor_connstrs)
-    log.info("postgres is running on 'test_restart_compute' branch")
+    print("postgres is running on 'test_restart_compute' branch")

    with closing(pg.connect()) as conn:
        with conn.cursor() as cur:
@@ -37,7 +39,7 @@ def test_restart_compute(
            cur.execute('SELECT sum(key) FROM t')
            r = cur.fetchone()
            assert r == (5000050000, )
-            log.info(f"res = {r}")
+            print("res = ", r)

    # Remove data directory and restart
    pg.stop_and_destroy().create_start('test_restart_compute',
@@ -50,7 +52,7 @@ def test_restart_compute(
            cur.execute('SELECT sum(key) FROM t')
            r = cur.fetchone()
            assert r == (5000050000, )
-            log.info(f"res = {r}")
+            print("res = ", r)

            # Insert another row
            cur.execute("INSERT INTO t VALUES (100001, 'payload2')")
@@ -58,7 +60,7 @@ def test_restart_compute(

            r = cur.fetchone()
            assert r == (100001, )
-            log.info(f"res = {r}")
+            print("res = ", r)

    # Again remove data directory and restart
    pg.stop_and_destroy().create_start('test_restart_compute',
@@ -73,7 +75,7 @@ def test_restart_compute(

            r = cur.fetchone()
            assert r == (100001, )
-            log.info(f"res = {r}")
+            print("res = ", r)

    # And again remove data directory and restart
    pg.stop_and_destroy().create_start('test_restart_compute',
@@ -86,4 +88,4 @@ def test_restart_compute(

            r = cur.fetchone()
            assert r == (100001, )
-            log.info(f"res = {r}")
+            print("res = ", r)
--- a/test_runner/batch_others/test_snapfiles_gc.py
+++ b/test_runner/batch_others/test_snapfiles_gc.py
@@ -1,14 +1,13 @@
 from contextlib import closing
 import psycopg2.extras
-import time
-from fixtures.log_helper import log
+import time;

 pytest_plugins = ("fixtures.zenith_fixtures")

 def print_gc_result(row):
-    log.info("GC duration {elapsed} ms".format_map(row));
-    log.info("  REL    total: {layer_relfiles_total}, needed_by_cutoff {layer_relfiles_needed_by_cutoff}, needed_by_branches: {layer_relfiles_needed_by_branches}, not_updated: {layer_relfiles_not_updated}, needed_as_tombstone {layer_relfiles_needed_as_tombstone}, removed: {layer_relfiles_removed}, dropped: {layer_relfiles_dropped}".format_map(row))
-    log.info("  NONREL total: {layer_nonrelfiles_total}, needed_by_cutoff {layer_nonrelfiles_needed_by_cutoff}, needed_by_branches: {layer_nonrelfiles_needed_by_branches}, not_updated: {layer_nonrelfiles_not_updated}, needed_as_tombstone {layer_nonrelfiles_needed_as_tombstone}, removed: {layer_nonrelfiles_removed}, dropped: {layer_nonrelfiles_dropped}".format_map(row))
+    print("GC duration {elapsed} ms".format_map(row));
+    print("  REL    total: {layer_relfiles_total}, needed_by_cutoff {layer_relfiles_needed_by_cutoff}, needed_by_branches: {layer_relfiles_needed_by_branches}, not_updated: {layer_relfiles_not_updated}, removed: {layer_relfiles_removed}, dropped: {layer_relfiles_dropped}".format_map(row))
+    print("  NONREL total: {layer_nonrelfiles_total}, needed_by_cutoff {layer_nonrelfiles_needed_by_cutoff}, needed_by_branches: {layer_nonrelfiles_needed_by_branches}, not_updated: {layer_nonrelfiles_not_updated}, removed: {layer_nonrelfiles_removed}, dropped: {layer_nonrelfiles_dropped}".format_map(row))


 #
@@ -36,7 +35,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):

                    cur.execute("select relfilenode from pg_class where oid = 'foo'::regclass");
                    row = cur.fetchone();
-                    log.info(f"relfilenode is {row[0]}");
+                    print("relfilenode is {}", row[0]);

                    # Run GC, to clear out any garbage left behind in the catalogs by
                    # the CREATE TABLE command. We want to have a clean slate with no garbage
@@ -51,7 +50,7 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                    # update to confuse our numbers either.
                    cur.execute("DELETE FROM foo")

-                    log.info("Running GC before test")
+                    print("Running GC before test")
                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                    row = pscur.fetchone()
                    print_gc_result(row);
@@ -59,46 +58,44 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                    layer_relfiles_remain = row['layer_relfiles_total'] - row['layer_relfiles_removed']
                    assert layer_relfiles_remain > 0

-                    # Insert a row and run GC. Checkpoint should freeze the layer
-                    # so that there is only the most recent image layer left for the rel,
-                    # removing the old image and delta layer.
-                    log.info("Inserting one row and running GC")
+                    # Insert a row.
+                    print("Inserting one row and running GC")
                    cur.execute("INSERT INTO foo VALUES (1)")
                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                    row = pscur.fetchone()
                    print_gc_result(row);
-                    assert row['layer_relfiles_total'] == layer_relfiles_remain + 2
-                    assert row['layer_relfiles_removed'] == 2
+                    assert row['layer_relfiles_total'] == layer_relfiles_remain + 1
+                    assert row['layer_relfiles_removed'] == 1
                    assert row['layer_relfiles_dropped'] == 0

                    # Insert two more rows and run GC.
-                    # This should create new image and delta layer file with the new contents, and
-                    # then remove the old one image and the just-created delta layer.
-                    log.info("Inserting two more rows and running GC")
+                    # This should create a new layer file with the new contents, and
+                    # remove the old one.
+                    print("Inserting two more rows and running GC")
                    cur.execute("INSERT INTO foo VALUES (2)")
                    cur.execute("INSERT INTO foo VALUES (3)")

                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                    row = pscur.fetchone()
                    print_gc_result(row);
-                    assert row['layer_relfiles_total'] == layer_relfiles_remain + 2
-                    assert row['layer_relfiles_removed'] == 2
+                    assert row['layer_relfiles_total'] == layer_relfiles_remain + 1
+                    assert row['layer_relfiles_removed'] == 1
                    assert row['layer_relfiles_dropped'] == 0

-                    # Do it again. Should again create two new layer files and remove old ones.
-                    log.info("Inserting two more rows and running GC")
+                    # Do it again. Should again create a new layer file and remove old one.
+                    print("Inserting two more rows and running GC")
                    cur.execute("INSERT INTO foo VALUES (2)")
                    cur.execute("INSERT INTO foo VALUES (3)")

                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                    row = pscur.fetchone()
                    print_gc_result(row);
-                    assert row['layer_relfiles_total'] == layer_relfiles_remain + 2
-                    assert row['layer_relfiles_removed'] == 2
+                    assert row['layer_relfiles_total'] == layer_relfiles_remain + 1
+                    assert row['layer_relfiles_removed'] == 1
                    assert row['layer_relfiles_dropped'] == 0

                    # Run GC again, with no changes in the database. Should not remove anything.
-                    log.info("Run GC again, with nothing to do")
+                    print("Run GC again, with nothing to do")
                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                    row = pscur.fetchone()
                    print_gc_result(row);
@@ -109,26 +106,19 @@ def test_layerfiles_gc(zenith_cli, pageserver, postgres, pg_bin):
                    #
                    # Test DROP TABLE checks that relation data and metadata was deleted by GC from object storage
                    #
-                    log.info("Drop table and run GC again");
+                    print("Drop table and run GC again");
                    cur.execute("DROP TABLE foo")

                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
                    row = pscur.fetchone()
                    print_gc_result(row);

-                    # We still cannot remove the latest layers
-                    # because they serve as tombstones for earlier layers.
-                    assert row['layer_relfiles_dropped'] == 0
                    # Each relation fork is counted separately, hence 3.
-                    assert row['layer_relfiles_needed_as_tombstone'] == 3
+                    assert row['layer_relfiles_dropped'] == 3

                    # The catalog updates also create new layer files of the catalogs, which
                    # are counted as 'removed'
                    assert row['layer_relfiles_removed'] > 0

-                    # TODO Change the test to check actual CG of dropped layers.
-                    # Each relation fork is counted separately, hence 3.
-                    #assert row['layer_relfiles_dropped'] == 3
-
                    # TODO: perhaps we should count catalog and user relations separately,
                    # to make this kind of testing more robust
--- a/test_runner/batch_others/test_timeline_size.py
+++ b/test_runner/batch_others/test_timeline_size.py
@@ -2,7 +2,7 @@ from contextlib import closing
 from uuid import UUID
 import psycopg2.extras
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log
+

 def test_timeline_size(
    zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFactory, pg_bin
@@ -15,7 +15,7 @@ def test_timeline_size(
    assert res["current_logical_size"] == res["current_logical_size_non_incremental"]

    pgmain = postgres.create_start("test_timeline_size")
-    log.info("postgres is running on 'test_timeline_size' branch")
+    print("postgres is running on 'test_timeline_size' branch")

    with closing(pgmain.connect()) as conn:
        with conn.cursor() as cur:
--- a/test_runner/batch_others/test_twophase.py
+++ b/test_runner/batch_others/test_twophase.py
@@ -1,7 +1,7 @@
 import os

 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver, PgBin
-from fixtures.log_helper import log
+

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -13,7 +13,7 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
    zenith_cli.run(["branch", "test_twophase", "empty"])

    pg = postgres.create_start('test_twophase', config_lines=['max_prepared_transactions=5'])
-    log.info("postgres is running on 'test_twophase' branch")
+    print("postgres is running on 'test_twophase' branch")

    conn = pg.connect()
    cur = conn.cursor()
@@ -45,7 +45,7 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
    cur.execute('CHECKPOINT')

    twophase_files = os.listdir(pg.pg_twophase_dir_path())
-    log.info(twophase_files)
+    print(twophase_files)
    assert len(twophase_files) == 4

    cur.execute("COMMIT PREPARED 'insert_three'")
@@ -53,7 +53,7 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa
    cur.execute('CHECKPOINT')

    twophase_files = os.listdir(pg.pg_twophase_dir_path())
-    log.info(twophase_files)
+    print(twophase_files)
    assert len(twophase_files) == 2

    # Create a branch with the transaction in prepared state
@@ -67,7 +67,7 @@ def test_twophase(zenith_cli, pageserver: ZenithPageserver, postgres: PostgresFa

    # Check that we restored only needed twophase files
    twophase_files2 = os.listdir(pg2.pg_twophase_dir_path())
-    log.info(twophase_files2)
+    print(twophase_files2)
    assert twophase_files2.sort() == twophase_files.sort()

    conn2 = pg2.connect()
--- a/test_runner/batch_others/test_vm_bits.py
+++ b/test_runner/batch_others/test_vm_bits.py
@@ -1,5 +1,4 @@
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -12,7 +11,7 @@ def test_vm_bit_clear(pageserver: ZenithPageserver, postgres: PostgresFactory, p
    zenith_cli.run(["branch", "test_vm_bit_clear", "empty"])
    pg = postgres.create_start('test_vm_bit_clear')

-    log.info("postgres is running on 'test_vm_bit_clear' branch")
+    print("postgres is running on 'test_vm_bit_clear' branch")
    pg_conn = pg.connect()
    cur = pg_conn.cursor()

@@ -64,7 +63,7 @@ def test_vm_bit_clear(pageserver: ZenithPageserver, postgres: PostgresFactory, p
    # server at the right point-in-time avoids that full-page image.
    pg_new = postgres.create_start('test_vm_bit_clear_new')

-    log.info("postgres is running on 'test_vm_bit_clear_new' branch")
+    print("postgres is running on 'test_vm_bit_clear_new' branch")
    pg_new_conn = pg_new.connect()
    cur_new = pg_new_conn.cursor()

--- a/test_runner/batch_others/test_wal_acceptor.py
+++ b/test_runner/batch_others/test_wal_acceptor.py
@@ -1,15 +1,10 @@
 import pytest
 import random
 import time
-import os
-import subprocess
-import uuid

 from contextlib import closing
 from multiprocessing import Process, Value
-from fixtures.zenith_fixtures import WalAcceptorFactory, ZenithPageserver, PostgresFactory, PgBin
-from fixtures.utils import lsn_to_hex, mkdir_if_needed
-from fixtures.log_helper import log
+from fixtures.zenith_fixtures import WalAcceptorFactory, ZenithPageserver, PostgresFactory

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -203,121 +198,3 @@ def test_race_conditions(zenith_cli, pageserver: ZenithPageserver, postgres: Pos

    stop_value.value = 1
    proc.join()
-
-
-class ProposerPostgres:
-    """Object for running safekeepers sync with walproposer"""
-    def __init__(self, pgdata_dir: str, pg_bin: PgBin, timeline_id: str, tenant_id: str):
-        self.pgdata_dir: str = pgdata_dir
-        self.pg_bin: PgBin = pg_bin
-        self.timeline_id: str = timeline_id
-        self.tenant_id: str = tenant_id
-
-    def pg_data_dir_path(self) -> str:
-        """ Path to data directory """
-        return self.pgdata_dir
-
-    def config_file_path(self) -> str:
-        """ Path to postgresql.conf """
-        return os.path.join(self.pgdata_dir, 'postgresql.conf')
-
-    def create_dir_config(self, wal_acceptors: str):
-        """ Create dir and config for running --sync-safekeepers """
-
-        mkdir_if_needed(self.pg_data_dir_path())
-        with open(self.config_file_path(), "w") as f:
-            f.writelines([
-                "synchronous_standby_names = 'walproposer'\n",
-                f"zenith.zenith_timeline = '{self.timeline_id}'\n",
-                f"zenith.zenith_tenant = '{self.tenant_id}'\n",
-                f"wal_acceptors = '{wal_acceptors}'\n",
-            ])
-
-    def sync_safekeepers(self) -> str:
-        """
-        Run 'postgres --sync-safekeepers'.
-        Returns execution result, which is commit_lsn after sync.
-        """
-
-        command = ["postgres", "--sync-safekeepers"]
-        env = {
-            "PGDATA": self.pg_data_dir_path(),
-        }
-
-        basepath = self.pg_bin.run_capture(command, env)
-        stdout_filename = basepath + '.stdout'
-
-        with open(stdout_filename, 'r') as stdout_f:
-            stdout = stdout_f.read()
-            return stdout.strip("\n ")
-
-
-# insert wal in all safekeepers and run sync on proposer
-def test_sync_safekeepers(repo_dir: str, pg_bin: PgBin, wa_factory: WalAcceptorFactory):
-    wa_factory.start_n_new(3)
-
-    timeline_id = uuid.uuid4().hex
-    tenant_id = uuid.uuid4().hex
-
-    # write config for proposer
-    pgdata_dir = os.path.join(repo_dir, "proposer_pgdata")
-    pg = ProposerPostgres(pgdata_dir, pg_bin, timeline_id, tenant_id)
-    pg.create_dir_config(wa_factory.get_connstrs())
-
-    # valid lsn, which is not in the segment start, nor in zero segment
-    epoch_start_lsn = 0x16B9188  # 0/16B9188
-    begin_lsn = epoch_start_lsn
-
-    # append and commit WAL
-    lsn_after_append = []
-    for i in range(3):
-        res = wa_factory.instances[i].append_logical_message(
-            tenant_id,
-            timeline_id,
-            {
-                "lm_prefix": "prefix",
-                "lm_message": "message",
-                "set_commit_lsn": True,
-                "term": 2,
-                "begin_lsn": begin_lsn,
-                "epoch_start_lsn": epoch_start_lsn,
-                "truncate_lsn": epoch_start_lsn,
-            },
-        )
-        lsn_hex = lsn_to_hex(res["inserted_wal"]["end_lsn"])
-        lsn_after_append.append(lsn_hex)
-        log.info(f"safekeeper[{i}] lsn after append: {lsn_hex}")
-
-    # run sync safekeepers
-    lsn_after_sync = pg.sync_safekeepers()
-    log.info(f"lsn after sync = {lsn_after_sync}")
-
-    assert all(lsn_after_sync == lsn for lsn in lsn_after_append)
-
-
-def test_timeline_status(zenith_cli, pageserver, postgres, wa_factory: WalAcceptorFactory):
-    wa_factory.start_n_new(1)
-
-    zenith_cli.run(["branch", "test_timeline_status", "empty"])
-    pg = postgres.create_start('test_timeline_status', wal_acceptors=wa_factory.get_connstrs())
-
-    wa = wa_factory.instances[0]
-    wa_http_cli = wa.http_client()
-    wa_http_cli.check_status()
-
-    # learn zenith timeline from compute
-    tenant_id = pg.safe_psql("show zenith.zenith_tenant")[0][0]
-    timeline_id = pg.safe_psql("show zenith.zenith_timeline")[0][0]
-
-    # fetch something sensible from status
-    epoch = wa_http_cli.timeline_status(tenant_id, timeline_id).acceptor_epoch
-
-    pg.safe_psql("create table t(i int)")
-
-    # ensure epoch goes up after reboot
-    pg.stop().start()
-    pg.safe_psql("insert into t values(10)")
-
-    epoch_after_reboot = wa_http_cli.timeline_status(tenant_id,
-                                                     timeline_id).acceptor_epoch
-    assert epoch_after_reboot > epoch
--- a/test_runner/batch_others/test_wal_acceptor_async.py
+++ b/test_runner/batch_others/test_wal_acceptor_async.py
@@ -3,10 +3,9 @@ import asyncpg
 import random

 from fixtures.zenith_fixtures import WalAcceptor, WalAcceptorFactory, ZenithPageserver, PostgresFactory, Postgres
-from fixtures.log_helper import getLogger
 from typing import List
+from fixtures.utils import debug_print

-log = getLogger('root.wal_acceptor_async')
 pytest_plugins = ("fixtures.zenith_fixtures")


@@ -64,18 +63,18 @@ class WorkerStats(object):
        self.counters[worker_id] += 1

    def check_progress(self):
-        log.debug("Workers progress: {}".format(self.counters))
+        debug_print("Workers progress: {}".format(self.counters))

        # every worker should finish at least one tx
        assert all(cnt > 0 for cnt in self.counters)

        progress = sum(self.counters)
-        log.info('All workers made {} transactions'.format(progress))
+        print('All workers made {} transactions'.format(progress))


 async def run_random_worker(stats: WorkerStats, pg: Postgres, worker_id, n_accounts, max_transfer):
    pg_conn = await pg.connect_async()
-    log.debug('Started worker {}'.format(worker_id))
+    debug_print('Started worker {}'.format(worker_id))

    while stats.running:
        from_uid = random.randint(0, n_accounts - 1)
@@ -85,9 +84,9 @@ async def run_random_worker(stats: WorkerStats, pg: Postgres, worker_id, n_accou
        await bank_transfer(pg_conn, from_uid, to_uid, amount)
        stats.inc_progress(worker_id)

-        log.debug('Executed transfer({}) {} => {}'.format(amount, from_uid, to_uid))
+        debug_print('Executed transfer({}) {} => {}'.format(amount, from_uid, to_uid))

-    log.debug('Finished worker {}'.format(worker_id))
+    debug_print('Finished worker {}'.format(worker_id))

    await pg_conn.close()

@@ -119,14 +118,9 @@ async def run_restarts_under_load(pg: Postgres, acceptors: List[WalAcceptor], n_
        victim = acceptors[it % len(acceptors)]
        victim.stop()

-        # Wait till previous victim recovers so it is ready for the next
-        # iteration by making any writing xact.
-        conn = await pg.connect_async()
-        await conn.execute(
-            'UPDATE bank_accs SET amount = amount WHERE uid = 1',
-            timeout=120
-        )
-        await conn.close()
+        # wait for transactions that could have started and finished before
+        # victim acceptor was stopped
+        await asyncio.sleep(1)

        stats.reset()
        await asyncio.sleep(period_time)
@@ -135,7 +129,7 @@ async def run_restarts_under_load(pg: Postgres, acceptors: List[WalAcceptor], n_

        victim.start()

-    log.info('Iterations are finished, exiting coroutines...')
+    print('Iterations are finished, exiting coroutines...')
    stats.running = False
    # await all workers
    await asyncio.gather(*workers)
--- a/test_runner/batch_pg_regress/test_zenith_regress.py
+++ b/test_runner/batch_pg_regress/test_zenith_regress.py
@@ -2,7 +2,6 @@ import os

 from fixtures.utils import mkdir_if_needed
 from fixtures.zenith_fixtures import PageserverPort, PostgresFactory, check_restored_datadir_content
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures")

@@ -39,7 +38,7 @@ def test_zenith_regress(postgres: PostgresFactory, pg_bin, zenith_cli, test_outp
        '--inputdir={}'.format(src_path),
    ]

-    log.info(pg_regress_command)
+    print(pg_regress_command)
    env = {
        'PGPORT': str(pg.port),
        'PGUSER': pg.username,
--- a/test_runner/fixtures/benchmark_fixture.py
+++ b/test_runner/fixtures/benchmark_fixture.py
@@ -1,3 +1,5 @@
+from pprint import pprint
+
 import os
 import re
 import timeit
@@ -134,34 +136,9 @@ class ZenithBenchmarker:
        # The metric should be an integer, as it's a number of bytes. But in general
        # all prometheus metrics are floats. So to be pedantic, read it as a float
        # and round to integer.
-        matches = re.search(r'^pageserver_disk_io_bytes{io_operation="write"} (\S+)$', all_metrics,
-                            re.MULTILINE)
+        matches = re.search(r'pageserver_disk_io_bytes{io_operation="write"} (\S+)', all_metrics)
        return int(round(float(matches.group(1))))

-    def get_peak_mem(self, pageserver) -> int:
-        """
-        Fetch the "maxrss" metric from the pageserver
-        """
-        # Fetch all the exposed prometheus metrics from page server
-        all_metrics = pageserver.http_client().get_metrics()
-        # See comment in get_io_writes()
-        matches = re.search(r'^pageserver_maxrss_kb (\S+)$', all_metrics,
-                            re.MULTILINE)
-        return int(round(float(matches.group(1))))
-
-    def get_timeline_size(self, repo_dir: str, tenantid: str, timelineid: str):
-        """
-        Calculate the on-disk size of a timeline
-        """
-        path = "{}/tenants/{}/timelines/{}".format(repo_dir, tenantid, timelineid)
-
-        totalbytes = 0
-        for root, dirs, files in os.walk(path):
-            for name in files:
-                totalbytes += os.path.getsize(os.path.join(root, name))
-
-        return totalbytes
-
    @contextmanager
    def record_pageserver_writes(self, pageserver, metric_name):
        """
--- a/test_runner/fixtures/log_helper.py
+++ b/test_runner/fixtures/log_helper.py
@@ -1,44 +0,0 @@
-import logging
-import logging.config
-
-"""
-This file configures logging to use in python tests.
-Logs are automatically captured and shown in their
-own section after all tests are executed. 
-
-To see logs for all (even successful) tests, run
-pytest with the following command:
- `pipenv run pytest -n8 -rA`
-
-Other log config can be set in pytest.ini file.
-You can add `log_cli = true` to it to watch
-logs in real time.
-
-To get more info about logging with pytest, see
-https://docs.pytest.org/en/6.2.x/logging.html
-"""
-
-# this config is only used for default log levels,
-# log format is specified in pytest.ini file
-LOGGING = {
-    "version": 1,
-    "loggers": {
-        "root": {
-            "level": "INFO"
-        },
-        "root.wal_acceptor_async": {
-            "level": "INFO" # a lot of logs on DEBUG level
-        }
-    }
-}
-
-def getLogger(name='root') -> logging.Logger:
-    """Method to get logger for tests.
-    
-    Should be used to get correctly initialized logger. """
-    return logging.getLogger(name)
-
-# default logger for tests
-log = getLogger()
-
-logging.config.dictConfig(LOGGING)
--- a/test_runner/fixtures/utils.py
+++ b/test_runner/fixtures/utils.py
@@ -2,7 +2,7 @@ import os
 import subprocess

 from typing import Any, List
-from fixtures.log_helper import log
+

 def get_self_dir() -> str:
    """ Get the path to the directory where this script lives. """
@@ -21,7 +21,7 @@ def mkdir_if_needed(path: str) -> None:
    assert os.path.isdir(path)


-def subprocess_capture(capture_dir: str, cmd: List[str], **kwargs: Any) -> str:
+def subprocess_capture(capture_dir: str, cmd: List[str], **kwargs: Any) -> None:
    """ Run a process and capture its output

    Output will go to files named "cmd_NNN.stdout" and "cmd_NNN.stderr"
@@ -29,7 +29,6 @@ def subprocess_capture(capture_dir: str, cmd: List[str], **kwargs: Any) -> str:
    counter.

    If those files already exist, we will overwrite them.
-    Returns basepath for files with captured output.
    """
    assert type(cmd) is list
    base = os.path.basename(cmd[0]) + '_{}'.format(global_counter())
@@ -39,11 +38,9 @@ def subprocess_capture(capture_dir: str, cmd: List[str], **kwargs: Any) -> str:

    with open(stdout_filename, 'w') as stdout_f:
        with open(stderr_filename, 'w') as stderr_f:
-            log.info('(capturing output to "{}.stdout")'.format(base))
+            print('(capturing output to "{}.stdout")'.format(base))
            subprocess.run(cmd, **kwargs, stdout=stdout_f, stderr=stderr_f)

-    return basepath
-

 _global_counter = 0

@@ -58,6 +55,10 @@ def global_counter() -> int:
    _global_counter += 1
    return _global_counter

-def lsn_to_hex(num: int) -> str:
-    """ Convert lsn from int to standard hex notation. """
-    return "{:X}/{:X}".format(num >> 32, num & 0xffffffff)
+def debug_print(*args, **kwargs) -> None:
+    """ Print to the console if TEST_DEBUG_PRINT is set in env.
+    
+    All parameters are passed to print().
+    """
+    if os.environ.get('TEST_DEBUG_PRINT') is not None:
+        print(*args, **kwargs)
--- a/test_runner/fixtures/zenith_fixtures.py
+++ b/test_runner/fixtures/zenith_fixtures.py
@@ -5,7 +5,6 @@ import os
 import pathlib
 import uuid
 import jwt
-import json
 import psycopg2
 import pytest
 import shutil
@@ -27,8 +26,6 @@ from typing_extensions import Literal
 import requests

 from .utils import (get_self_dir, mkdir_if_needed, subprocess_capture)
-from fixtures.log_helper import log
-
 """
 This file contains pytest fixtures. A fixture is a test resource that can be
 summoned by placing its name in the test's arguments.
@@ -66,16 +63,14 @@ def pytest_configure(config):
         raise Exception('Too many workers configured. Cannot distrubute ports for services.')

    # does not use -c as it is not supported on macOS
-    cmd = ['pgrep', 'pageserver|postgres|safekeeper']
+    cmd = ['pgrep', 'pageserver|postgres|wal_acceptor']
    result = subprocess.run(cmd, stdout=subprocess.DEVNULL)
    if result.returncode == 0:
        # returncode of 0 means it found something.
        # This is bad; we don't want any of those processes polluting the
        # result of the test.
        # NOTE this shows as an internal pytest error, there might be a better way
-        raise Exception(
-            'Found interfering processes running. Stop all Zenith pageservers, nodes, safekeepers, as well as stand-alone Postgres.'
-        )
+        raise Exception('found interfering processes running')


 def determine_scope(fixture_name: str, config: Any) -> str:
@@ -190,13 +185,13 @@ class ZenithCli:

        >>> result = zenith_cli.run(...)
        >>> assert result.stderr == ""
-        >>> log.info(result.stdout)
+        >>> print(result.stdout)
        """

        assert type(arguments) == list

        args = [self.bin_zenith] + arguments
-        log.info('Running command "{}"'.format(' '.join(args)))
+        print('Running command "{}"'.format(' '.join(args)))

        # Interceipt CalledProcessError and print more info
        try:
@@ -213,7 +208,7 @@ class ZenithCli:
              stdout: {exc.stdout}
              stderr: {exc.stderr}
            """
-            log.info(msg)
+            print(msg)

            raise Exception(msg) from exc

@@ -235,16 +230,16 @@ class ZenithPageserverHttpClient(requests.Session):
            self.headers['Authorization'] = f'Bearer {auth_token}'

    def check_status(self):
-        self.get(f"http://127.0.0.1:{self.port}/v1/status").raise_for_status()
+        self.get(f"http://localhost:{self.port}/v1/status").raise_for_status()

    def branch_list(self, tenant_id: uuid.UUID) -> List[Dict]:
-        res = self.get(f"http://127.0.0.1:{self.port}/v1/branch/{tenant_id.hex}")
+        res = self.get(f"http://localhost:{self.port}/v1/branch/{tenant_id.hex}")
        res.raise_for_status()
        return res.json()

    def branch_create(self, tenant_id: uuid.UUID, name: str, start_point: str) -> Dict:
        res = self.post(
-            f"http://127.0.0.1:{self.port}/v1/branch",
+            f"http://localhost:{self.port}/v1/branch",
            json={
                'tenant_id': tenant_id.hex,
                'name': name,
@@ -256,19 +251,19 @@ class ZenithPageserverHttpClient(requests.Session):

    def branch_detail(self, tenant_id: uuid.UUID, name: str) -> Dict:
        res = self.get(
-            f"http://127.0.0.1:{self.port}/v1/branch/{tenant_id.hex}/{name}",
+            f"http://localhost:{self.port}/v1/branch/{tenant_id.hex}/{name}",
        )
        res.raise_for_status()
        return res.json()

    def tenant_list(self) -> List[str]:
-        res = self.get(f"http://127.0.0.1:{self.port}/v1/tenant")
+        res = self.get(f"http://localhost:{self.port}/v1/tenant")
        res.raise_for_status()
        return res.json()

    def tenant_create(self, tenant_id: uuid.UUID):
        res = self.post(
-            f"http://127.0.0.1:{self.port}/v1/tenant",
+            f"http://localhost:{self.port}/v1/tenant",
            json={
                'tenant_id': tenant_id.hex,
            },
@@ -277,7 +272,7 @@ class ZenithPageserverHttpClient(requests.Session):
        return res.json()

    def get_metrics(self) -> str:
-        res = self.get(f"http://127.0.0.1:{self.port}/metrics")
+        res = self.get(f"http://localhost:{self.port}/metrics")
        res.raise_for_status()
        return res.text

@@ -348,7 +343,7 @@ class PageserverPort:
 class ZenithPageserver(PgProtocol):
    """ An object representing a running pageserver. """
    def __init__(self, zenith_cli: ZenithCli, repo_dir: str, port: PageserverPort):
-        super().__init__(host='127.0.0.1', port=port.pg)
+        super().__init__(host='localhost', port=port.pg)
        self.zenith_cli = zenith_cli
        self.running = False
        self.initial_tenant = None
@@ -377,7 +372,6 @@ class ZenithPageserver(PgProtocol):
        Start the page server.
        Returns self.
        """
-        assert self.running == False

        self.zenith_cli.run(['start'])
        self.running = True
@@ -385,18 +379,14 @@ class ZenithPageserver(PgProtocol):
        self.initial_tenant = self.zenith_cli.run(['tenant', 'list']).stdout.strip()
        return self

-    def stop(self, immediate=False) -> 'ZenithPageserver':
+    def stop(self) -> 'ZenithPageserver':
        """
        Stop the page server.
        Returns self.
        """
-        cmd = ['stop']
-        if immediate:
-            cmd.append('immediate')

-        log.info(f"Stopping pageserver with {cmd}")
        if self.running:
-            self.zenith_cli.run(cmd)
+            self.zenith_cli.run(['stop'])
            self.running = False

        return self
@@ -405,7 +395,7 @@ class ZenithPageserver(PgProtocol):
        return self

    def __exit__(self, exc_type, exc, tb):
-        self.stop(True)
+        self.stop()

    @cached_property
    def auth_keys(self) -> AuthKeys:
@@ -426,7 +416,7 @@ class ZenithPageserver(PgProtocol):
 def pageserver_port(port_distributor: PortDistributor) -> PageserverPort:
    pg = port_distributor.get_port()
    http = port_distributor.get_port()
-    log.info(f"pageserver_port: pg={pg} http={http}")
+    print(f"pageserver_port: pg={pg} http={http}")
    return PageserverPort(pg=pg, http=http)


@@ -450,8 +440,8 @@ def pageserver(zenith_cli: ZenithCli, repo_dir: str, pageserver_port: Pageserver
    yield ps

    # After the yield comes any cleanup code we need.
-    log.info('Starting pageserver cleanup')
-    ps.stop(True)
+    print('Starting pageserver cleanup')
+    ps.stop()

 class PgBin:
    """ A helper class for executing postgres binaries """
@@ -488,26 +478,24 @@ class PgBin:
        """

        self._fixpath(command)
-        log.info('Running command "{}"'.format(' '.join(command)))
+        print('Running command "{}"'.format(' '.join(command)))
        env = self._build_env(env)
        subprocess.run(command, env=env, cwd=cwd, check=True)

    def run_capture(self,
                    command: List[str],
                    env: Optional[Env] = None,
-                    cwd: Optional[str] = None,
-                    **kwargs: Any) -> None:
+                    cwd: Optional[str] = None) -> None:
        """
        Run one of the postgres binaries, with stderr and stdout redirected to a file.

-        This is just like `run`, but for chatty programs. Returns basepath for files
-        with captured output.
+        This is just like `run`, but for chatty programs.
        """

        self._fixpath(command)
-        log.info('Running command "{}"'.format(' '.join(command)))
+        print('Running command "{}"'.format(' '.join(command)))
        env = self._build_env(env)
-        return subprocess_capture(self.log_dir, command, env=env, cwd=cwd, check=True, **kwargs)
+        subprocess_capture(self.log_dir, command, env=env, cwd=cwd, check=True)


@zenfixture
@@ -525,7 +513,7 @@ def pageserver_auth_enabled(zenith_cli: ZenithCli, repo_dir: str, pageserver_por
 class Postgres(PgProtocol):
    """ An object representing a running postgres daemon. """
    def __init__(self, zenith_cli: ZenithCli, repo_dir: str, pg_bin: PgBin, tenant_id: str, port: int):
-        super().__init__(host='127.0.0.1', port=port)
+        super().__init__(host='localhost', port=port)

        self.zenith_cli = zenith_cli
        self.running = False
@@ -573,12 +561,12 @@ class Postgres(PgProtocol):

        assert self.branch is not None

-        log.info(f"Starting postgres on branch {self.branch}")
+        print(f"Starting postgres on branch {self.branch}")

        run_result = self.zenith_cli.run(['pg', 'start', self.branch, f'--tenantid={self.tenant_id}', f'--port={self.port}'])
        self.running = True

-        log.info(f"stdout: {run_result.stdout}")
+        print(f"stdout: {run_result.stdout}")

        return self

@@ -793,7 +781,7 @@ def postgres(zenith_cli: ZenithCli, initial_tenant: str, repo_dir: str, pg_bin:
    yield pgfactory

    # After the yield comes any cleanup code we need.
-    log.info('Starting postgres cleanup')
+    print('Starting postgres cleanup')
    pgfactory.stop_all()

 def read_pid(path: Path):
@@ -801,17 +789,12 @@ def read_pid(path: Path):
    return int(path.read_text())


-@dataclass
-class WalAcceptorPort:
-    pg: int
-    http: int
-
@dataclass
 class WalAcceptor:
    """ An object representing a running wal acceptor daemon. """
    wa_bin_path: Path
    data_dir: Path
-    port: WalAcceptorPort
+    port: int
    num: int # identifier for logging
    pageserver_port: int
    auth_token: Optional[str] = None
@@ -823,36 +806,28 @@ class WalAcceptor:

        cmd = [str(self.wa_bin_path)]
        cmd.extend(["-D", str(self.data_dir)])
-        cmd.extend(["--listen-pg", f"127.0.0.1:{self.port.pg}"])
-        cmd.extend(["--listen-http", f"127.0.0.1:{self.port.http}"])
+        cmd.extend(["-l", f"localhost:{self.port}"])
        cmd.append("--daemonize")
        cmd.append("--no-sync")
        # Tell page server it can receive WAL from this WAL safekeeper
-        cmd.extend(["--pageserver", f"127.0.0.1:{self.pageserver_port}"])
+        cmd.extend(["--pageserver", f"localhost:{self.pageserver_port}"])
        cmd.extend(["--recall", "1 second"])
-        log.info('Running command "{}"'.format(' '.join(cmd)))
+        print('Running command "{}"'.format(' '.join(cmd)))
        env = {'PAGESERVER_AUTH_TOKEN': self.auth_token} if self.auth_token else None
        subprocess.run(cmd, check=True, env=env)

-        # wait for wal acceptor start by checking its status
-        started_at = time.time()
-        while True:
-            try:
-                http_cli = self.http_client()
-                http_cli.check_status()
-            except Exception as e:
-                elapsed = time.time() - started_at
-                if elapsed > 3:
-                    raise RuntimeError(
-                        f"timed out waiting {elapsed:.0f}s for wal acceptor start: {e}")
-                time.sleep(0.5)
-            else:
-                break  # success
-        return self
+        # wait for wal acceptor start by checkking that pid is readable
+        for _ in range(3):
+            pid = self.get_pid()
+            if pid is not None:
+                return self
+            time.sleep(0.5)
+
+        raise RuntimeError("cannot get wal acceptor pid")

    @property
    def pidfile(self) -> Path:
-        return self.data_dir / "safekeeper.pid"
+        return self.data_dir / "wal_acceptor.pid"

    def get_pid(self) -> Optional[int]:
        if not self.pidfile.exists():
@@ -866,10 +841,10 @@ class WalAcceptor:
        return pid

    def stop(self) -> 'WalAcceptor':
-        log.info('Stopping wal acceptor {}'.format(self.num))
+        print('Stopping wal acceptor {}'.format(self.num))
        pid = self.get_pid()
        if pid is None:
-            log.info("Wal acceptor {} is not running".format(self.num))
+            print("Wal acceptor {} is not running".format(self.num))
            return self

        try:
@@ -879,36 +854,11 @@ class WalAcceptor:
            pass # pidfile might be obsolete
        return self

-    def append_logical_message(self, tenant_id: str, timeline_id: str, request: Dict[str, Any]) -> Dict[str, Any]:
-        """
-        Send JSON_CTRL query to append LogicalMessage to WAL and modify 
-        safekeeper state. It will construct LogicalMessage from provided
-        prefix and message, and then will write it to WAL.
-        """
-
-        # "replication=0" hacks psycopg not to send additional queries
-        # on startup, see https://github.com/psycopg/psycopg2/pull/482
-        connstr = f"host=127.0.0.1 port={self.port.pg} replication=0 options='-c ztimelineid={timeline_id} ztenantid={tenant_id}'"
-
-        with closing(psycopg2.connect(connstr)) as conn:
-            # server doesn't support transactions
-            conn.autocommit = True
-            with conn.cursor() as cur:
-                request_json = json.dumps(request)
-                log.info(f"JSON_CTRL request on port {self.port.pg}: {request_json}")
-                cur.execute("JSON_CTRL " + request_json)
-                all = cur.fetchall()
-                log.info(f"JSON_CTRL response: {all[0][0]}")
-                return json.loads(all[0][0])
-
-    def http_client(self):
-        return WalAcceptorHttpClient(port=self.port.http)
-

 class WalAcceptorFactory:
    """ An object representing multiple running wal acceptors. """
    def __init__(self, zenith_binpath: Path, data_dir: Path, pageserver_port: int, port_distributor: PortDistributor):
-        self.wa_bin_path = zenith_binpath / 'safekeeper'
+        self.wa_bin_path = zenith_binpath / 'wal_acceptor'
        self.data_dir = data_dir
        self.instances: List[WalAcceptor] = []
        self.port_distributor = port_distributor
@@ -922,10 +872,7 @@ class WalAcceptorFactory:
        wa = WalAcceptor(
            wa_bin_path=self.wa_bin_path,
            data_dir=self.data_dir / "wal_acceptor_{}".format(wa_num),
-            port=WalAcceptorPort(
-                pg=self.port_distributor.get_port(),
-                http=self.port_distributor.get_port(),
-            ),
+            port=self.port_distributor.get_port(),
            num=wa_num,
            pageserver_port=self.pageserver_port,
            auth_token=auth_token,
@@ -949,7 +896,7 @@ class WalAcceptorFactory:

    def get_connstrs(self) -> str:
        """ Get list of wal acceptor endpoints suitable for wal_acceptors GUC  """
-        return ','.join(["127.0.0.1:{}".format(wa.port.pg) for wa in self.instances])
+        return ','.join(["localhost:{}".format(wa.port) for wa in self.instances])


@zenfixture
@@ -963,34 +910,16 @@ def wa_factory(zenith_binpath: str, repo_dir: str, pageserver_port: PageserverPo
    )
    yield wafactory
    # After the yield comes any cleanup code we need.
-    log.info('Starting wal acceptors cleanup')
+    print('Starting wal acceptors cleanup')
    wafactory.stop_all()

-@dataclass
-class PageserverTimelineStatus:
-    acceptor_epoch: int
-
-class WalAcceptorHttpClient(requests.Session):
-    def __init__(self, port: int) -> None:
-        super().__init__()
-        self.port = port
-
-    def check_status(self):
-        self.get(f"http://127.0.0.1:{self.port}/v1/status").raise_for_status()
-
-    def timeline_status(self, tenant_id: str, timeline_id: str) -> PageserverTimelineStatus:
-        res = self.get(f"http://127.0.0.1:{self.port}/v1/timeline/{tenant_id}/{timeline_id}")
-        res.raise_for_status()
-        resj = res.json()
-        return PageserverTimelineStatus(acceptor_epoch=resj['acceptor_state']['epoch'])
-

@zenfixture
 def base_dir() -> str:
    """ find the base directory (currently this is the git root) """

    base_dir = os.path.normpath(os.path.join(get_self_dir(), '../..'))
-    log.info(f'base_dir is {base_dir}')
+    print('\nbase_dir is', base_dir)
    return base_dir


@@ -1019,7 +948,7 @@ def test_output_dir(request: Any, top_output_dir: str) -> str:
        test_name = 'shared'

    test_output_dir = os.path.join(top_output_dir, test_name)
-    log.info(f'test_output_dir is {test_output_dir}')
+    print('test_output_dir is', test_output_dir)
    shutil.rmtree(test_output_dir, ignore_errors=True)
    mkdir_if_needed(test_output_dir)
    return test_output_dir
@@ -1061,7 +990,7 @@ def pg_distrib_dir(base_dir: str) -> str:
        pg_dir = env_postgres_bin
    else:
        pg_dir = os.path.normpath(os.path.join(base_dir, DEFAULT_POSTGRES_DIR))
-    log.info(f'postgres dir is {pg_dir}')
+    print('postgres dir is', pg_dir)
    if not os.path.exists(os.path.join(pg_dir, 'bin/postgres')):
        raise Exception('postgres not found at "{}"'.format(pg_dir))
    return pg_dir
@@ -1100,7 +1029,7 @@ def list_files_to_compare(pgdata_dir: str):
                pgdata_files.append(rel_file)

    pgdata_files.sort()
-    log.info(pgdata_files)
+    print(pgdata_files)
    return pgdata_files

 # pg is the existing and running compute node, that we want to compare with a basebackup
@@ -1124,7 +1053,7 @@ def check_restored_datadir_content(zenith_cli: ZenithCli, test_output_dir: str,
    cmd = rf"""
        {psql_path}                                    \
            --no-psqlrc                                \
-            postgres://127.0.0.1:{pageserver_pg_port}  \
+            postgres://localhost:{pageserver_pg_port}  \
            -c 'basebackup {pg.tenant_id} {timeline}'  \
         | tar -x -C {restored_dir_path}
    """
@@ -1146,7 +1075,9 @@ def check_restored_datadir_content(zenith_cli: ZenithCli, test_output_dir: str,
                                                restored_dir_path,
                                                pgdata_files,
                                                shallow=False)
-    log.info(f'filecmp result mismatch and error lists:\n\t mismatch={mismatch}\n\t error={error}')
+    print('filecmp result mismatch and error lists:')
+    print(mismatch)
+    print(error)

    for f in mismatch:

--- a/test_runner/netstat-script.sh
+++ b/test_runner/netstat-script.sh
@@ -1,8 +0,0 @@
-#!/bin/bash
-
-while true; do
-  echo -n "==== CURRENT TIME:" >> /tmp/test_output/netstat.stdout
-  date +"%T.%N" >> /tmp/test_output/netstat.stdout
-  sudo netstat -vpnoa | grep tcp | sort >> /tmp/test_output/netstat.stdout
-  sleep 0.5
-done
--- a/test_runner/performance/test_bulk_insert.py
+++ b/test_runner/performance/test_bulk_insert.py
@@ -1,10 +1,22 @@
 import os
 from contextlib import closing
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")

+def get_timeline_size(repo_dir: str, tenantid: str, timelineid: str):
+    path = "{}/tenants/{}/timelines/{}".format(repo_dir, tenantid, timelineid)
+
+    totalbytes = 0
+    for root, dirs, files in os.walk(path):
+        for name in files:
+            totalbytes += os.path.getsize(os.path.join(root, name))
+
+        if 'wal' in dirs:
+            dirs.remove('wal')  # don't visit 'wal' subdirectory
+
+    return totalbytes
+
 #
 # Run bulk INSERT test.
 #
@@ -13,14 +25,13 @@ pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
 # 1. Time to INSERT 5 million rows
 # 2. Disk writes
 # 3. Disk space used
-# 4. Peak memory usage
 #
 def test_bulk_insert(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin, zenith_cli, zenbenchmark, repo_dir: str):
    # Create a branch for us
    zenith_cli.run(["branch", "test_bulk_insert", "empty"])

    pg = postgres.create_start('test_bulk_insert')
-    log.info("postgres is running on 'test_bulk_insert' branch")
+    print("postgres is running on 'test_bulk_insert' branch")

    # Open a connection directly to the page server that we'll use to force
    # flushing the layers to disk
@@ -44,9 +55,6 @@ def test_bulk_insert(postgres: PostgresFactory, pageserver: ZenithPageserver, pg
                    # time and I/O
                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")

-            # Record peak memory usage
-            zenbenchmark.record("peak_mem", zenbenchmark.get_peak_mem(pageserver) / 1024, 'MB')
-
            # Report disk space used by the repository
-            timeline_size = zenbenchmark.get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
+            timeline_size = get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
            zenbenchmark.record('size', timeline_size / (1024*1024), 'MB')
--- a/test_runner/performance/test_bulk_tenant_create.py
+++ b/test_runner/performance/test_bulk_tenant_create.py
@@ -1,58 +0,0 @@
-import timeit
-import pytest
-
-from fixtures.zenith_fixtures import (
-    TenantFactory,
-    ZenithCli,
-    PostgresFactory,
-)
-
-pytest_plugins = ("fixtures.benchmark_fixture")
-
-# Run bulk tenant creation test.
-#
-# Collects metrics:
-#
-# 1. Time to create {1,10,50} tenants
-# 2. Average creation time per tenant
-
-
-@pytest.mark.parametrize('tenants_count', [1, 5, 10])
-@pytest.mark.parametrize('use_wal_acceptors', ['with_wa', 'without_wa'])
-def test_bulk_tenant_create(
-    zenith_cli: ZenithCli,
-    tenant_factory: TenantFactory,
-    postgres: PostgresFactory,
-    wa_factory,
-    use_wal_acceptors: str,
-    tenants_count: int,
-    zenbenchmark,
-):
-    """Measure tenant creation time (with and without wal acceptors)"""
-
-    time_slices = []
-
-    for i in range(tenants_count):
-        start = timeit.default_timer()
-
-        tenant = tenant_factory.create()
-        zenith_cli.run([
-            "branch", f"test_bulk_tenant_create_{tenants_count}_{i}_{use_wal_acceptors}", "main",
-            f"--tenantid={tenant}"
-        ])
-
-        if use_wal_acceptors == 'with_wa':
-            wa_factory.start_n_new(3)
-
-        pg_tenant = postgres.create_start(
-            f"test_bulk_tenant_create_{tenants_count}_{i}_{use_wal_acceptors}",
-            tenant,
-            wal_acceptors=wa_factory.get_connstrs() if use_wal_acceptors == 'with_wa' else None,
-        )
-
-        end = timeit.default_timer()
-        time_slices.append(end - start)
-
-        pg_tenant.stop()
-
-    zenbenchmark.record('tenant_creation_time', sum(time_slices) / len(time_slices), 's')
--- a/test_runner/performance/test_gist_build.py
+++ b/test_runner/performance/test_gist_build.py
@@ -1,49 +0,0 @@
-import os
-from contextlib import closing
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log
-
-pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
-
-#
-# Test buffering GisT build. It WAL-logs the whole relation, in 32-page chunks.
-# As of this writing, we're duplicate those giant WAL records for each page,
-# which makes the delta layer about 32x larger than it needs to be.
-#
-def test_gist_buffering_build(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin, zenith_cli, zenbenchmark, repo_dir: str):
-    # Create a branch for us
-    zenith_cli.run(["branch", "test_gist_buffering_build", "empty"])
-
-    pg = postgres.create_start('test_gist_buffering_build')
-    log.info("postgres is running on 'test_gist_buffering_build' branch")
-
-    # Open a connection directly to the page server that we'll use to force
-    # flushing the layers to disk
-    psconn = pageserver.connect();
-    pscur = psconn.cursor()
-
-    # Get the timeline ID of our branch. We need it for the 'do_gc' command
-    with closing(pg.connect()) as conn:
-        with conn.cursor() as cur:
-            cur.execute("SHOW zenith.zenith_timeline")
-            timeline = cur.fetchone()[0]
-
-            # Create test table.
-            cur.execute("create table gist_point_tbl(id int4, p point)");
-            cur.execute("insert into gist_point_tbl select g, point(g, g) from generate_series(1, 1000000) g;");
-
-            # Build the index.
-            with zenbenchmark.record_pageserver_writes(pageserver, 'pageserver_writes'):
-                with zenbenchmark.record_duration('build'):
-                    cur.execute("create index gist_pointidx2 on gist_point_tbl using gist(p) with (buffering = on)");
-
-                    # Flush the layers from memory to disk. This is included in the reported
-                    # time and I/O
-                    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 1000000")
-
-            # Record peak memory usage
-            zenbenchmark.record("peak_mem", zenbenchmark.get_peak_mem(pageserver) / 1024, 'MB')
-
-            # Report disk space used by the repository
-            timeline_size = zenbenchmark.get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
-            zenbenchmark.record('size', timeline_size / (1024*1024), 'MB')
--- a/test_runner/performance/test_perf_pgbench.py
+++ b/test_runner/performance/test_perf_pgbench.py
@@ -1,10 +1,22 @@
 import os
 from contextlib import closing
 from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log

 pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")

+def get_timeline_size(repo_dir: str, tenantid: str, timelineid: str):
+    path = "{}/tenants/{}/timelines/{}".format(repo_dir, tenantid, timelineid)
+
+    totalbytes = 0
+    for root, dirs, files in os.walk(path):
+        for name in files:
+            totalbytes += os.path.getsize(os.path.join(root, name))
+
+        if 'wal' in dirs:
+            dirs.remove('wal')  # don't visit 'wal' subdirectory
+
+    return totalbytes
+
 #
 # Run a very short pgbench test.
 #
@@ -19,7 +31,7 @@ def test_pgbench(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin
    zenith_cli.run(["branch", "test_pgbench_perf", "empty"])

    pg = postgres.create_start('test_pgbench_perf')
-    log.info("postgres is running on 'test_pgbench_perf' branch")
+    print("postgres is running on 'test_pgbench_perf' branch")

    # Open a connection directly to the page server that we'll use to force
    # flushing the layers to disk
@@ -52,5 +64,5 @@ def test_pgbench(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin
    pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")

    # Report disk space used by the repository
-    timeline_size = zenbenchmark.get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
+    timeline_size = get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
    zenbenchmark.record('size', timeline_size / (1024*1024), 'MB')
--- a/test_runner/performance/test_write_amplification.py
+++ b/test_runner/performance/test_write_amplification.py
@@ -1,75 +0,0 @@
-# Demonstrate Write Amplification with naive oldest-first layer checkpointing
-# algorithm.
-#
-# In each iteration of the test, we create a new table that's slightly under 10
-# MB in size (10 MB is the current "segment size" used by the page server). Then
-# we make a tiny update to all the tables already created. This creates a WAL
-# pattern where you have a lot of updates on one segment (the newly created
-# one), alternating with a small updates on all relations. This is the worst
-# case scenario for the naive checkpointing policy where we write out the layers
-# in LSN order, writing the oldest layer first. That creates a new 10 MB image
-# layer to be created for each of those small updates.  This is the Write
-# Amplification problem at its finest.
-import os
-from contextlib import closing
-from fixtures.zenith_fixtures import PostgresFactory, ZenithPageserver
-from fixtures.log_helper import log
-
-pytest_plugins = ("fixtures.zenith_fixtures", "fixtures.benchmark_fixture")
-
-def test_write_amplification(postgres: PostgresFactory, pageserver: ZenithPageserver, pg_bin, zenith_cli, zenbenchmark, repo_dir: str):
-    # Create a branch for us
-    zenith_cli.run(["branch", "test_write_amplification", "empty"])
-
-    pg = postgres.create_start('test_write_amplification')
-    log.info("postgres is running on 'test_write_amplification' branch")
-
-    # Open a connection directly to the page server that we'll use to force
-    # flushing the layers to disk
-    psconn = pageserver.connect();
-    pscur = psconn.cursor()
-
-    with closing(pg.connect()) as conn:
-        with conn.cursor() as cur:
-            # Get the timeline ID of our branch. We need it for the 'do_gc' command
-            cur.execute("SHOW zenith.zenith_timeline")
-            timeline = cur.fetchone()[0]
-
-            with zenbenchmark.record_pageserver_writes(pageserver, 'pageserver_writes'):
-                with zenbenchmark.record_duration('run'):
-
-                    # NOTE: Because each iteration updates every table already created,
-                    # the runtime and write amplification is O(n^2), where n is the
-                    # number of iterations.
-                    for i in range(25):
-                        cur.execute(f'''
-                        CREATE TABLE tbl{i} AS
-                            SELECT g as i, 'long string to consume some space' || g as t
-                            FROM generate_series(1, 100000) g
-                        ''')
-                        cur.execute(f"create index on tbl{i} (i);")
-                        for j in range(1, i):
-                            cur.execute(f"delete from tbl{j} where i = {i}")
-
-                        # Force checkpointing. As of this writing, we don't have
-                        # a back-pressure mechanism, and the page server cannot
-                        # keep up digesting and checkpointing the WAL at the
-                        # rate that it is generated. If we don't force a
-                        # checkpoint, the WAL will just accumulate in memory
-                        # until you hit OOM error. So in effect, we use much
-                        # more memory to hold the incoming WAL, and write them
-                        # out in larger batches than we'd really want. Using
-                        # more memory hides the write amplification problem this
-                        # test tries to demonstrate.
-                        #
-                        # The write amplification problem is real, and using
-                        # more memory isn't the right solution. We could
-                        # demonstrate the effect also by generating the WAL
-                        # slower, adding some delays in this loop.  But forcing
-                        # the the checkpointing and GC makes the test go faster,
-                        # with the same total I/O effect.
-                        pscur.execute(f"do_gc {pageserver.initial_tenant} {timeline} 0")
-
-            # Report disk space used by the repository
-            timeline_size = zenbenchmark.get_timeline_size(repo_dir, pageserver.initial_tenant, timeline)
-            zenbenchmark.record('size', timeline_size / (1024*1024), 'MB')
--- a/test_runner/pytest.ini
+++ b/test_runner/pytest.ini
@@ -1,4 +1,2 @@
 [pytest]
 minversion = 6.0
-log_format = %(asctime)s.%(msecs)-3d %(levelname)s [%(filename)s:%(lineno)d] %(message)s
-log_date_format = %Y-%m-%d %H:%M:%S
--- a/test_runner/test_broken.py
+++ b/test_runner/test_broken.py
@@ -1,8 +1,6 @@
 import pytest
 import os

-from fixtures.log_helper import log
-
 pytest_plugins = ("fixtures.zenith_fixtures")
 """
 Use this test to see what happens when tests fail.
@@ -24,7 +22,7 @@ def test_broken(zenith_cli, pageserver, postgres, pg_bin):
    zenith_cli.run(["branch", "test_broken", "empty"])

    postgres.create_start("test_broken")
-    log.info('postgres is running')
+    print('postgres is running')

-    log.info('THIS NEXT COMMAND WILL FAIL:')
+    print('THIS NEXT COMMAND WILL FAIL:')
    pg_bin.run('pgbench -i_am_a_broken_test'.split())
--- a/vendor/postgres
+++ b/vendor/postgres
--- a/walkeeper/Cargo.toml
+++ b/walkeeper/Cargo.toml
@@ -11,11 +11,8 @@ regex = "1.4.5"
 bincode = "1.3"
 bytes = "1.0.1"
 byteorder = "1.4.3"
-hyper = "0.14"
-routerify = "2"
 fs2 = "0.4.3"
 lazy_static = "1.4.0"
-serde_json = "1"
 log = "0.4.14"
 clap = "2.33.0"
 daemonize = "0.4.1"
@@ -30,11 +27,9 @@ humantime = "2.1.0"
 walkdir = "2"
 serde = { version = "1.0", features = ["derive"] }
 hex = "0.4.3"
-const_format = "0.2.21"

 # FIXME: 'pageserver' is needed for ZTimelineId. Refactor
 pageserver = { path = "../pageserver" }
 postgres_ffi = { path = "../postgres_ffi" }
 workspace_hack = { path = "../workspace_hack" }
-zenith_metrics = { path = "../zenith_metrics" }
 zenith_utils = { path = "../zenith_utils" }
--- a/walkeeper/README
+++ b/walkeeper/README
@@ -76,43 +76,6 @@ safekeepers.
 See README_PROTO.md for a more detailed desription of the consensus
 protocol. spec/ contains TLA+ specification of it.

-# Q&A
-
-Q: Why have a separate service instead of connecting Page Server directly to a
-   primary PostgreSQL node?
-A: Page Server is a single server which can be lost. As our primary
-   fault-tolerant storage is S3, we do not want to wait for it before
-   committing a transaction. The WAL service acts as a temporary fault-tolerant
-   storage for recent data before it gets to the Page Server and then finally
-   to S3. Whenever WALs and pages are committed to S3, WAL's storage can be
-   trimmed.
-
-Q: What if the compute node evicts a page, needs it back, but the page is yet
-   to reach the Page Server?
-A: If the compute node has evicted a page, changes to it have been WAL-logged
-   (that's why it is called Write Ahead logging; there are some exceptions like
-   index builds, but these are exceptions). These WAL records will eventually
-   reach the Page Server. The Page Server notes that the compute note requests
-   pages with a very recent LSN and will not respond to the compute node until a
-   corresponding WAL is received from WAL safekeepers.
-
-Q: How long may Page Server wait for?
-A: Not too long, hopefully. If a page is evicted, it probably was not used for
-   a while, so the WAL service have had enough time to push changes to the Page
-   Server. There may be issues if there is no backpressure and compute node with
-   WAL service run ahead of Page Server, though.
-   There is no backpressure right now, so you may even see some spurious
-   timeouts in tests.
-
-Q: How do WAL safekeepers communicate with each other?
-A: They may only send each other messages via the compute node, they never
-   communicate directly with each other.
-
-Q: Why have a consensus algorithm if there is only a single compute node?
-A: Actually there may be moments with multiple PostgreSQL nodes running at the
-   same time. E.g. we are bringing one up and one down. We would like to avoid
-   simultaneous writes from different nodes, so there should be a consensus on
-   who is the primary node.

 # Terminology

--- a/walkeeper/src/bin/wal_acceptor.rs
+++ b/walkeeper/src/bin/wal_acceptor.rs
@@ -1,48 +1,35 @@
 //
-// Main entry point for the safekeeper executable
+// Main entry point for the wal_acceptor executable
 //
 use anyhow::Result;
 use clap::{App, Arg};
-use const_format::formatcp;
 use daemonize::Daemonize;
 use log::*;
 use std::env;
-use std::net::TcpListener;
 use std::path::{Path, PathBuf};
 use std::thread;
-use zenith_utils::http::endpoint;
 use zenith_utils::logging;

-use walkeeper::defaults::{DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_PG_LISTEN_ADDR};
-use walkeeper::http;
 use walkeeper::s3_offload;
 use walkeeper::wal_service;
 use walkeeper::WalAcceptorConf;

 fn main() -> Result<()> {
-    zenith_metrics::set_common_metrics_prefix("safekeeper");
-    let arg_matches = App::new("Zenith safekeeper")
+    let arg_matches = App::new("Zenith wal_acceptor")
        .about("Store WAL stream to local file system and push it to WAL receivers")
        .arg(
            Arg::with_name("datadir")
                .short("D")
                .long("dir")
                .takes_value(true)
-                .help("Path to the safekeeper data directory"),
+                .help("Path to the WAL acceptor data directory"),
        )
        .arg(
-            Arg::with_name("listen-pg")
+            Arg::with_name("listen")
                .short("l")
-                .long("listen-pg")
-                .alias("listen") // for compatibility
+                .long("listen")
                .takes_value(true)
-                .help(formatcp!("listen for incoming WAL data connections on ip:port (default: {DEFAULT_PG_LISTEN_ADDR})")),
-        )
-        .arg(
-            Arg::with_name("listen-http")
-                .long("listen-http")
-                .takes_value(true)
-                .help(formatcp!("http endpoint address for metrics on ip:port (default: {DEFAULT_HTTP_LISTEN_ADDR})")),
+                .help("listen for incoming connections on ip:port (default: 127.0.0.1:5454)"),
        )
        .arg(
            Arg::with_name("pageserver")
@@ -83,8 +70,7 @@ fn main() -> Result<()> {
        daemonize: false,
        no_sync: false,
        pageserver_addr: None,
-        listen_pg_addr: DEFAULT_PG_LISTEN_ADDR.to_string(),
-        listen_http_addr: DEFAULT_HTTP_LISTEN_ADDR.to_string(),
+        listen_addr: "localhost:5454".to_string(),
        ttl: None,
        recall_period: None,
        pageserver_auth_token: env::var("PAGESERVER_AUTH_TOKEN").ok(),
@@ -105,12 +91,8 @@ fn main() -> Result<()> {
        conf.daemonize = true;
    }

-    if let Some(addr) = arg_matches.value_of("listen-pg") {
-        conf.listen_pg_addr = addr.to_owned();
-    }
-
-    if let Some(addr) = arg_matches.value_of("listen-http") {
-        conf.listen_http_addr = addr.to_owned();
+    if let Some(addr) = arg_matches.value_of("listen") {
+        conf.listen_addr = addr.to_owned();
    }

    if let Some(addr) = arg_matches.value_of("pageserver") {
@@ -129,19 +111,8 @@ fn main() -> Result<()> {
 }

 fn start_wal_acceptor(conf: WalAcceptorConf) -> Result<()> {
-    let log_filename = conf.data_dir.join("safekeeper.log");
-    let log_file = logging::init(log_filename, conf.daemonize)?;
-
-    let http_listener = TcpListener::bind(conf.listen_http_addr.clone()).map_err(|e| {
-        error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
-        e
-    })?;
-
-    info!("Starting safekeeper on {}", conf.listen_pg_addr);
-    let pg_listener = TcpListener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
-        error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
-        e
-    })?;
+    let log_filename = conf.data_dir.join("wal_acceptor.log");
+    let (_scope_guard, log_file) = logging::init(log_filename, conf.daemonize)?;

    if conf.daemonize {
        info!("daemonizing...");
@@ -152,7 +123,7 @@ fn start_wal_acceptor(conf: WalAcceptorConf) -> Result<()> {
        let stderr = log_file;

        let daemonize = Daemonize::new()
-            .pid_file("safekeeper.pid")
+            .pid_file("wal_acceptor.pid")
            .working_directory(Path::new("."))
            .stdout(stdout)
            .stderr(stderr);
@@ -165,17 +136,6 @@ fn start_wal_acceptor(conf: WalAcceptorConf) -> Result<()> {

    let mut threads = Vec::new();

-    let conf_cloned = conf.clone();
-    let http_endpoint_thread = thread::Builder::new()
-        .name("http_endpoint_thread".into())
-        .spawn(|| {
-            // TODO authentication
-            let router = http::make_router(conf_cloned);
-            endpoint::serve_thread_main(router, http_listener).unwrap();
-        })
-        .unwrap();
-    threads.push(http_endpoint_thread);
-
    if conf.ttl.is_some() {
        let s3_conf = conf.clone();
        let s3_offload_thread = thread::Builder::new()
@@ -192,7 +152,7 @@ fn start_wal_acceptor(conf: WalAcceptorConf) -> Result<()> {
        .name("WAL acceptor thread".into())
        .spawn(|| {
            // thread code
-            let thread_result = wal_service::thread_main(conf, pg_listener);
+            let thread_result = wal_service::thread_main(conf);
            if let Err(e) = thread_result {
                info!("wal_service thread terminated: {}", e);
            }
--- a/walkeeper/src/http/mod.rs
+++ b/walkeeper/src/http/mod.rs
@@ -1,2 +0,0 @@
-pub mod routes;
-pub use routes::make_router;
--- a/walkeeper/src/http/routes.rs
+++ b/walkeeper/src/http/routes.rs
@@ -1,88 +0,0 @@
-use hyper::{Body, Request, Response, StatusCode};
-use routerify::ext::RequestExt;
-use routerify::RouterBuilder;
-use serde::Serialize;
-use serde::Serializer;
-use std::fmt::Display;
-use std::sync::Arc;
-use zenith_utils::lsn::Lsn;
-
-use crate::safekeeper::AcceptorState;
-use crate::timeline::CreateControlFile;
-use crate::timeline::GlobalTimelines;
-use crate::WalAcceptorConf;
-use zenith_utils::http::endpoint;
-use zenith_utils::http::error::ApiError;
-use zenith_utils::http::json::json_response;
-use zenith_utils::http::request::parse_request_param;
-use zenith_utils::zid::{ZTenantId, ZTimelineId};
-
-/// Healthcheck handler.
-async fn status_handler(_: Request<Body>) -> Result<Response<Body>, ApiError> {
-    Ok(json_response(StatusCode::OK, "")?)
-}
-
-fn get_conf(request: &Request<Body>) -> &WalAcceptorConf {
-    request
-        .data::<Arc<WalAcceptorConf>>()
-        .expect("unknown state type")
-        .as_ref()
-}
-
-fn display_serialize<S, F>(z: &F, s: S) -> Result<S::Ok, S::Error>
-where
-    S: Serializer,
-    F: Display,
-{
-    s.serialize_str(&format!("{}", z))
-}
-
-/// Info about timeline on safekeeper ready for reporting.
-#[derive(Debug, Serialize)]
-struct TimelineStatus {
-    #[serde(serialize_with = "display_serialize")]
-    tenant_id: ZTenantId,
-    #[serde(serialize_with = "display_serialize")]
-    timeline_id: ZTimelineId,
-    acceptor_state: AcceptorState,
-    #[serde(serialize_with = "display_serialize")]
-    commit_lsn: Lsn,
-    #[serde(serialize_with = "display_serialize")]
-    truncate_lsn: Lsn,
-}
-
-/// Report info about timeline.
-async fn timeline_status_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
-    let tenant_id: ZTenantId = parse_request_param(&request, "tenant_id")?;
-    let timeline_id: ZTimelineId = parse_request_param(&request, "timeline_id")?;
-
-    let tli = GlobalTimelines::get(
-        get_conf(&request),
-        tenant_id,
-        timeline_id,
-        CreateControlFile::False,
-    )
-    .map_err(ApiError::from_err)?;
-    let sk_state = tli.get_info();
-
-    let status = TimelineStatus {
-        tenant_id,
-        timeline_id,
-        acceptor_state: sk_state.acceptor_state,
-        commit_lsn: sk_state.commit_lsn,
-        truncate_lsn: sk_state.truncate_lsn,
-    };
-    Ok(json_response(StatusCode::OK, status)?)
-}
-
-/// Safekeeper http router.
-pub fn make_router(conf: WalAcceptorConf) -> RouterBuilder<hyper::Body, ApiError> {
-    let router = endpoint::make_router();
-    router
-        .data(Arc::new(conf))
-        .get("/v1/status", status_handler)
-        .get(
-            "/v1/timeline/:tenant_id/:timeline_id",
-            timeline_status_handler,
-        )
-}
--- a/walkeeper/src/json_ctrl.rs
+++ b/walkeeper/src/json_ctrl.rs
@@ -1,249 +0,0 @@
-//!
-//! This module implements JSON_CTRL protocol, which allows exchange
-//! JSON messages over psql for testing purposes.
-//!
-//! Currently supports AppendLogicalMessage, which is used for WAL
-//! modifications in tests.
-//!
-
-use anyhow::{anyhow, Result};
-use bytes::{BufMut, Bytes, BytesMut};
-use crc32c::crc32c_append;
-use log::*;
-use serde::{Deserialize, Serialize};
-
-use crate::safekeeper::{AcceptorProposerMessage, AppendResponse};
-use crate::safekeeper::{
-    AppendRequest, AppendRequestHeader, ProposerAcceptorMessage, ProposerGreeting,
-};
-use crate::safekeeper::{SafeKeeperState, Term};
-use crate::send_wal::SendWalHandler;
-use crate::timeline::TimelineTools;
-use postgres_ffi::pg_constants;
-use postgres_ffi::xlog_utils;
-use postgres_ffi::{uint32, uint64, Oid, XLogRecord};
-use zenith_utils::lsn::Lsn;
-use zenith_utils::postgres_backend::PostgresBackend;
-use zenith_utils::pq_proto::{BeMessage, RowDescriptor, TEXT_OID};
-
-#[derive(Serialize, Deserialize, Debug)]
-struct AppendLogicalMessage {
-    // prefix and message to build LogicalMessage
-    lm_prefix: String,
-    lm_message: String,
-
-    // if true, commit_lsn will match flush_lsn after append
-    set_commit_lsn: bool,
-
-    // fields from AppendRequestHeader
-    term: Term,
-    epoch_start_lsn: Lsn,
-    begin_lsn: Lsn,
-    truncate_lsn: Lsn,
-}
-
-#[derive(Serialize, Deserialize)]
-struct AppendResult {
-    // safekeeper state after append
-    state: SafeKeeperState,
-    // info about new record in the WAL
-    inserted_wal: InsertedWAL,
-}
-
-/// Handles command to craft logical message WAL record with given
-/// content, and then append it with specified term and lsn. This
-/// function is used to test safekeepers in different scenarios.
-pub fn handle_json_ctrl(
-    swh: &mut SendWalHandler,
-    pgb: &mut PostgresBackend,
-    cmd: &Bytes,
-) -> Result<()> {
-    let cmd = cmd
-        .strip_prefix(b"JSON_CTRL")
-        .ok_or_else(|| anyhow!("invalid prefix"))?;
-    // trim zeroes in the end
-    let cmd = cmd.strip_suffix(&[0u8]).unwrap_or(cmd);
-
-    let append_request: AppendLogicalMessage = serde_json::from_slice(cmd)?;
-    info!("JSON_CTRL request: {:?}", append_request);
-
-    // need to init safekeeper state before AppendRequest
-    prepare_safekeeper(swh)?;
-
-    let inserted_wal = append_logical_message(swh, append_request)?;
-    let response = AppendResult {
-        state: swh.timeline.get().get_info(),
-        inserted_wal,
-    };
-    let response_data = serde_json::to_vec(&response)?;
-
-    pgb.write_message_noflush(&BeMessage::RowDescription(&[RowDescriptor {
-        name: b"json",
-        typoid: TEXT_OID,
-        typlen: -1,
-        ..Default::default()
-    }]))?
-    .write_message_noflush(&BeMessage::DataRow(&[Some(&response_data)]))?
-    .write_message(&BeMessage::CommandComplete(b"JSON_CTRL"))?;
-    Ok(())
-}
-
-/// Prepare safekeeper to process append requests without crashes,
-/// by sending ProposerGreeting with default server.wal_seg_size.
-fn prepare_safekeeper(swh: &mut SendWalHandler) -> Result<()> {
-    let greeting_request = ProposerAcceptorMessage::Greeting(ProposerGreeting {
-        protocol_version: 1, // current protocol
-        pg_version: 0,       // unknown
-        proposer_id: [0u8; 16],
-        system_id: 0,
-        ztli: swh.timelineid.unwrap(),
-        tenant_id: swh.tenantid.unwrap(),
-        tli: 0,
-        wal_seg_size: pg_constants::WAL_SEGMENT_SIZE as u32, // 16MB, default for tests
-    });
-
-    let response = swh.timeline.get().process_msg(&greeting_request)?;
-    match response {
-        AcceptorProposerMessage::Greeting(_) => Ok(()),
-        _ => anyhow::bail!("not GreetingResponse"),
-    }
-}
-
-#[derive(Serialize, Deserialize)]
-struct InsertedWAL {
-    begin_lsn: Lsn,
-    end_lsn: Lsn,
-    append_response: AppendResponse,
-}
-
-/// Extend local WAL with new LogicalMessage record. To do that,
-/// create AppendRequest with new WAL and pass it to safekeeper.
-fn append_logical_message(
-    swh: &mut SendWalHandler,
-    msg: AppendLogicalMessage,
-) -> Result<InsertedWAL> {
-    let wal_data = encode_logical_message(msg.lm_prefix, msg.lm_message);
-    let sk_state = swh.timeline.get().get_info();
-
-    let begin_lsn = msg.begin_lsn;
-    let end_lsn = begin_lsn + wal_data.len() as u64;
-
-    let commit_lsn = if msg.set_commit_lsn {
-        end_lsn
-    } else {
-        sk_state.commit_lsn
-    };
-
-    let append_request = ProposerAcceptorMessage::AppendRequest(AppendRequest {
-        h: AppendRequestHeader {
-            term: msg.term,
-            epoch_start_lsn: begin_lsn,
-            begin_lsn,
-            end_lsn,
-            commit_lsn,
-            truncate_lsn: msg.truncate_lsn,
-            proposer_uuid: [0u8; 16],
-        },
-        wal_data: Bytes::from(wal_data),
-    });
-
-    let response = swh.timeline.get().process_msg(&append_request)?;
-
-    let append_response = match response {
-        AcceptorProposerMessage::AppendResponse(resp) => resp,
-        _ => anyhow::bail!("not AppendResponse"),
-    };
-
-    Ok(InsertedWAL {
-        begin_lsn,
-        end_lsn,
-        append_response,
-    })
-}
-
-#[repr(C)]
-#[derive(Debug, Clone, Default, Serialize, Deserialize)]
-struct XlLogicalMessage {
-    db_id: Oid,
-    transactional: uint32, // bool, takes 4 bytes due to alignment in C structures
-    prefix_size: uint64,
-    message_size: uint64,
-}
-
-impl XlLogicalMessage {
-    pub fn encode(&self) -> Bytes {
-        use zenith_utils::bin_ser::LeSer;
-        self.ser().unwrap().into()
-    }
-}
-
-/// Create new WAL record for non-transactional logical message.
-/// Used for creating artificial WAL for tests, as LogicalMessage
-/// record is basically no-op.
-fn encode_logical_message(prefix: String, message: String) -> Vec<u8> {
-    let mut prefix_bytes = BytesMut::with_capacity(prefix.len() + 1);
-    prefix_bytes.put(prefix.as_bytes());
-    prefix_bytes.put_u8(0);
-
-    let message_bytes = message.as_bytes();
-
-    let logical_message = XlLogicalMessage {
-        db_id: 0,
-        transactional: 0,
-        prefix_size: prefix_bytes.len() as u64,
-        message_size: message_bytes.len() as u64,
-    };
-
-    let mainrdata = logical_message.encode();
-    let mainrdata_len: usize = mainrdata.len() + prefix_bytes.len() + message_bytes.len();
-    // only short mainrdata is supported for now
-    assert!(mainrdata_len <= 255);
-    let mainrdata_len = mainrdata_len as u8;
-
-    let mut data: Vec<u8> = vec![pg_constants::XLR_BLOCK_ID_DATA_SHORT, mainrdata_len];
-    data.extend_from_slice(&mainrdata);
-    data.extend_from_slice(&prefix_bytes);
-    data.extend_from_slice(message_bytes);
-
-    let total_len = xlog_utils::XLOG_SIZE_OF_XLOG_RECORD + data.len();
-
-    let mut header = XLogRecord {
-        xl_tot_len: total_len as u32,
-        xl_xid: 0,
-        xl_prev: 0,
-        xl_info: 0,
-        xl_rmid: 21,
-        __bindgen_padding_0: [0u8; 2usize],
-        xl_crc: 0, // crc will be calculated later
-    };
-
-    let header_bytes = header.encode();
-    let crc = crc32c_append(0, &data);
-    let crc = crc32c_append(crc, &header_bytes[0..xlog_utils::XLOG_RECORD_CRC_OFFS]);
-    header.xl_crc = crc;
-
-    let mut wal: Vec<u8> = Vec::new();
-    wal.extend_from_slice(&header.encode());
-    wal.extend_from_slice(&data);
-
-    // WAL start position must be aligned at 8 bytes,
-    // this will add padding for the next WAL record.
-    const PADDING: usize = 8;
-    let padding_rem = wal.len() % PADDING;
-    if padding_rem != 0 {
-        wal.resize(wal.len() + PADDING - padding_rem, 0);
-    }
-
-    wal
-}
-
-#[test]
-fn test_encode_logical_message() {
-    let expected = [
-        64, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 21, 0, 0, 170, 34, 166, 227, 255, 38,
-        0, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 112, 114, 101, 102,
-        105, 120, 0, 109, 101, 115, 115, 97, 103, 101,
-    ];
-    let actual = encode_logical_message("prefix".to_string(), "message".to_string());
-    assert_eq!(expected, actual[..]);
-}
--- a/walkeeper/src/lib.rs
+++ b/walkeeper/src/lib.rs
@@ -2,8 +2,6 @@
 use std::path::PathBuf;
 use std::time::Duration;

-pub mod http;
-pub mod json_ctrl;
 pub mod receive_wal;
 pub mod replication;
 pub mod s3_offload;
@@ -12,23 +10,12 @@ pub mod send_wal;
 pub mod timeline;
 pub mod wal_service;

-pub mod defaults {
-    use const_format::formatcp;
-
-    pub const DEFAULT_PG_LISTEN_PORT: u16 = 5454;
-    pub const DEFAULT_PG_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_PG_LISTEN_PORT}");
-
-    pub const DEFAULT_HTTP_LISTEN_PORT: u16 = 7676;
-    pub const DEFAULT_HTTP_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_HTTP_LISTEN_PORT}");
-}
-
 #[derive(Debug, Clone)]
 pub struct WalAcceptorConf {
    pub data_dir: PathBuf,
    pub daemonize: bool,
    pub no_sync: bool,
-    pub listen_pg_addr: String,
-    pub listen_http_addr: String,
+    pub listen_addr: String,
    pub pageserver_addr: Option<String>,
    // TODO (create issue) this is temporary, until protocol between PG<->SK<->PS rework
    pub pageserver_auth_token: Option<String>,
--- a/walkeeper/src/receive_wal.rs
+++ b/walkeeper/src/receive_wal.rs
@@ -42,7 +42,7 @@ fn request_callback(conf: WalAcceptorConf, timelineid: ZTimelineId, tenantid: ZT
    );

    // use Config parsing because SockAddr parsing doesnt allow to use host names instead of ip addresses
-    let me_connstr = format!("postgresql://no_user@{}/no_db", conf.listen_pg_addr);
+    let me_connstr = format!("postgresql://no_user@{}/no_db", conf.listen_addr);
    let me_conf: Config = me_connstr.parse().unwrap();
    let (host, port) = connection_host_port(&me_conf);
    let callme = format!(
--- a/walkeeper/src/safekeeper.rs
+++ b/walkeeper/src/safekeeper.rs
@@ -15,11 +15,8 @@ use std::cmp::min;
 use std::io;
 use std::io::Read;

-use lazy_static::lazy_static;
-
 use crate::replication::HotStandbyFeedback;
 use postgres_ffi::xlog_utils::MAX_SEND_SIZE;
-use zenith_metrics::{register_gauge_vec, Gauge, GaugeVec};
 use zenith_utils::bin_ser::LeSer;
 use zenith_utils::lsn::Lsn;
 use zenith_utils::pq_proto::SystemId;
@@ -31,7 +28,7 @@ const SK_PROTOCOL_VERSION: u32 = 1;
 const UNKNOWN_SERVER_VERSION: u32 = 0;

 /// Consensus logical timestamp.
-pub type Term = u64;
+type Term = u64;

 /// Unique id of proposer. Not needed for correctness, used for monitoring.
 type PgUuid = [u8; 16];
@@ -114,7 +111,7 @@ impl Default for SafeKeeperState {
 // protocol messages

 /// Initial Proposer -> Acceptor message
-#[derive(Debug, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct ProposerGreeting {
    /// proposer-acceptor protocol version
    pub protocol_version: u32,
@@ -131,19 +128,19 @@ pub struct ProposerGreeting {

 /// Acceptor -> Proposer initial response: the highest term known to me
 /// (acceptor voted for).
-#[derive(Debug, Serialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct AcceptorGreeting {
    term: u64,
 }

 /// Vote request sent from proposer to safekeepers
-#[derive(Debug, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct VoteRequest {
    term: Term,
 }

 /// Vote itself, sent from safekeeper to proposer
-#[derive(Debug, Serialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct VoteResponse {
    term: Term,      // not really needed, just a sanity check
    vote_given: u64, // fixme u64 due to padding
@@ -155,26 +152,26 @@ pub struct VoteResponse {

 /// Request with WAL message sent from proposer to safekeeper. Along the way it
 /// announces 1) successful election (with epoch_start_lsn); 2) commit_lsn.
-#[derive(Debug)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct AppendRequest {
-    pub h: AppendRequestHeader,
-    pub wal_data: Bytes,
+    h: AppendRequestHeader,
+    wal_data: Bytes,
 }
-#[derive(Debug, Clone, Deserialize)]
+#[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct AppendRequestHeader {
-    pub term: Term,
+    term: Term,
    // LSN since the proposer appends WAL; determines epoch switch point.
-    pub epoch_start_lsn: Lsn,
+    epoch_start_lsn: Lsn,
    /// start position of message in WAL
-    pub begin_lsn: Lsn,
+    begin_lsn: Lsn,
    /// end position of message in WAL
-    pub end_lsn: Lsn,
+    end_lsn: Lsn,
    /// LSN committed by quorum of safekeepers
-    pub commit_lsn: Lsn,
+    commit_lsn: Lsn,
    /// minimal LSN which may be needed by proposer to perform recovery of some safekeeper
-    pub truncate_lsn: Lsn,
+    truncate_lsn: Lsn,
    // only for logging/debugging
-    pub proposer_uuid: PgUuid,
+    proposer_uuid: PgUuid,
 }

 /// Report safekeeper state to proposer
@@ -278,47 +275,8 @@ impl AcceptorProposerMessage {
 pub trait Storage {
    /// Persist safekeeper state on disk, optionally syncing it.
    fn persist(&mut self, s: &SafeKeeperState, sync: bool) -> Result<()>;
-    /// Write piece of wal in buf to disk and sync it.
-    fn write_wal(&mut self, server: &ServerInfo, startpos: Lsn, buf: &[u8]) -> Result<()>;
-}
-
-lazy_static! {
-    // The prometheus crate does not support u64 yet, i64 only (see `IntGauge`).
-    // i64 is faster than f64, so update to u64 when available.
-    static ref FLUSH_LSN_GAUGE: GaugeVec = register_gauge_vec!(
-        "safekeeper_flush_lsn",
-        "Current flush_lsn, grouped by timeline",
-        &["ztli"]
-    )
-    .expect("Failed to register safekeeper_flush_lsn gauge vec");
-    static ref COMMIT_LSN_GAUGE: GaugeVec = register_gauge_vec!(
-        "safekeeper_commit_lsn",
-        "Current commit_lsn (not necessarily persisted to disk), grouped by timeline",
-        &["ztli"]
-    )
-    .expect("Failed to register safekeeper_commit_lsn gauge vec");
-}
-
-struct SafeKeeperMetrics {
-    flush_lsn: Gauge,
-    commit_lsn: Gauge,
-}
-
-impl SafeKeeperMetrics {
-    fn new(ztli: ZTimelineId) -> SafeKeeperMetrics {
-        let ztli_str = format!("{}", ztli);
-        SafeKeeperMetrics {
-            flush_lsn: FLUSH_LSN_GAUGE.with_label_values(&[&ztli_str]),
-            commit_lsn: COMMIT_LSN_GAUGE.with_label_values(&[&ztli_str]),
-        }
-    }
-
-    fn new_noname() -> SafeKeeperMetrics {
-        SafeKeeperMetrics {
-            flush_lsn: FLUSH_LSN_GAUGE.with_label_values(&["n/a"]),
-            commit_lsn: COMMIT_LSN_GAUGE.with_label_values(&["n/a"]),
-        }
-    }
+    /// Write piece of wal in buf to disk.
+    fn write_wal(&mut self, s: &SafeKeeperState, startpos: Lsn, buf: &[u8]) -> Result<()>;
 }

 /// SafeKeeper which consumes events (messages from compute) and provides
@@ -328,8 +286,6 @@ pub struct SafeKeeper<ST: Storage> {
    /// Established by reading wal.
    pub flush_lsn: Lsn,
    pub tli: u32,
-    // Cached metrics so we don't have to recompute labels on each update.
-    metrics: SafeKeeperMetrics,
    /// not-yet-flushed pairs of same named fields in s.*
    pub commit_lsn: Lsn,
    pub truncate_lsn: Lsn,
@@ -348,7 +304,6 @@ where
        SafeKeeper {
            flush_lsn,
            tli,
-            metrics: SafeKeeperMetrics::new_noname(),
            commit_lsn: state.commit_lsn,
            truncate_lsn: state.truncate_lsn,
            storage,
@@ -400,8 +355,6 @@ where
        self.s.server.wal_seg_size = msg.wal_seg_size;
        self.storage.persist(&self.s, true)?;

-        self.metrics = SafeKeeperMetrics::new(self.s.server.ztli);
-
        info!(
            "processed greeting from proposer {:?}, sending term {:?}",
            msg.proposer_id, self.s.acceptor_state.term
@@ -470,7 +423,7 @@ where
        let mut last_rec_lsn = Lsn(0);
        if !msg.wal_data.is_empty() {
            self.storage
-                .write_wal(&self.s.server, msg.h.begin_lsn, &msg.wal_data)?;
+                .write_wal(&self.s, msg.h.begin_lsn, &msg.wal_data)?;

            // figure out last record's end lsn for reporting (if we got the
            // whole record)
@@ -525,7 +478,6 @@ where
        }
        if last_rec_lsn > self.flush_lsn {
            self.flush_lsn = last_rec_lsn;
-            self.metrics.flush_lsn.set(u64::from(self.flush_lsn) as f64);
        }

        // Advance commit_lsn taking into account what we have locally. xxx this
@@ -543,9 +495,6 @@ where
            sync_control_file |=
                commit_lsn >= msg.h.epoch_start_lsn && self.s.commit_lsn < msg.h.epoch_start_lsn;
            self.commit_lsn = commit_lsn;
-            self.metrics
-                .commit_lsn
-                .set(u64::from(self.commit_lsn) as f64);
        }

        self.truncate_lsn = msg.h.truncate_lsn;
@@ -597,7 +546,7 @@ mod tests {
            Ok(())
        }

-        fn write_wal(&mut self, _server: &ServerInfo, _startpos: Lsn, _buf: &[u8]) -> Result<()> {
+        fn write_wal(&mut self, _s: &SafeKeeperState, _startpos: Lsn, _buf: &[u8]) -> Result<()> {
            Ok(())
        }
    }
--- a/walkeeper/src/send_wal.rs
+++ b/walkeeper/src/send_wal.rs
@@ -2,7 +2,6 @@
 //! pageserver/any other consumer.
 //!

-use crate::json_ctrl::handle_json_ctrl;
 use crate::receive_wal::ReceiveWalConn;
 use crate::replication::ReplicationConn;
 use crate::timeline::{Timeline, TimelineTools};
@@ -13,13 +12,14 @@ use std::str::FromStr;
 use std::sync::Arc;
 use zenith_utils::postgres_backend;
 use zenith_utils::postgres_backend::PostgresBackend;
-use zenith_utils::pq_proto::{BeMessage, FeStartupMessage, RowDescriptor, INT4_OID, TEXT_OID};
+use zenith_utils::pq_proto::{BeMessage, FeStartupMessage, RowDescriptor};
 use zenith_utils::zid::{ZTenantId, ZTimelineId};

 use crate::timeline::CreateControlFile;

 /// Handler for streaming WAL from acceptor
 pub struct SendWalHandler {
+    /// wal acceptor configuration
    pub conf: WalAcceptorConf,
    /// assigned application name
    pub appname: Option<String>,
@@ -49,11 +49,9 @@ impl postgres_backend::Handler for SendWalHandler {
    }

    fn process_query(&mut self, pgb: &mut PostgresBackend, query_string: Bytes) -> Result<()> {
-        // START_WAL_PUSH is the only command that initializes the timeline in production.
-        // There is also JSON_CTRL command, which should initialize the timeline for testing.
+        // START_WAL_PUSH is the only command that initializes the timeline
        if self.timeline.is_none() {
-            if query_string.starts_with(b"START_WAL_PUSH") || query_string.starts_with(b"JSON_CTRL")
-            {
+            if query_string.starts_with(b"START_WAL_PUSH") {
                self.timeline.set(
                    &self.conf,
                    self.tenantid.unwrap(),
@@ -71,16 +69,16 @@ impl postgres_backend::Handler for SendWalHandler {
        }
        if query_string.starts_with(b"IDENTIFY_SYSTEM") {
            self.handle_identify_system(pgb)?;
+            Ok(())
        } else if query_string.starts_with(b"START_REPLICATION") {
            ReplicationConn::new(pgb).run(self, pgb, &query_string)?;
+            Ok(())
        } else if query_string.starts_with(b"START_WAL_PUSH") {
            ReceiveWalConn::new(pgb)?.run(self)?;
-        } else if query_string.starts_with(b"JSON_CTRL") {
-            handle_json_ctrl(self, pgb, &query_string)?;
+            Ok(())
        } else {
            bail!("Unexpected command {:?}", query_string);
        }
-        Ok(())
    }
 }

@@ -110,25 +108,25 @@ impl SendWalHandler {
        pgb.write_message_noflush(&BeMessage::RowDescription(&[
            RowDescriptor {
                name: b"systemid",
-                typoid: TEXT_OID,
+                typoid: 25,
                typlen: -1,
                ..Default::default()
            },
            RowDescriptor {
                name: b"timeline",
-                typoid: INT4_OID,
+                typoid: 23,
                typlen: 4,
                ..Default::default()
            },
            RowDescriptor {
                name: b"xlogpos",
-                typoid: TEXT_OID,
+                typoid: 25,
                typlen: -1,
                ..Default::default()
            },
            RowDescriptor {
                name: b"dbname",
-                typoid: TEXT_OID,
+                typoid: 25,
                typlen: -1,
                ..Default::default()
            },
--- a/walkeeper/src/timeline.rs
+++ b/walkeeper/src/timeline.rs
@@ -18,8 +18,8 @@ use zenith_utils::zid::{ZTenantId, ZTimelineId};

 use crate::replication::{HotStandbyFeedback, END_REPLICATION_MARKER};
 use crate::safekeeper::{
-    AcceptorProposerMessage, ProposerAcceptorMessage, SafeKeeper, SafeKeeperState, ServerInfo,
-    Storage, SK_FORMAT_VERSION, SK_MAGIC,
+    AcceptorProposerMessage, ProposerAcceptorMessage, SafeKeeper, SafeKeeperState, Storage,
+    SK_FORMAT_VERSION, SK_MAGIC,
 };
 use crate::WalAcceptorConf;
 use postgres_ffi::xlog_utils::{XLogFileName, XLOG_BLCKSZ};
@@ -32,7 +32,7 @@ struct SharedState {
    sk: SafeKeeper<FileStorage>,
    /// For receiving-sending wal cooperation
    /// quorum commit LSN we've notified walsenders about
-    notified_commit_lsn: Lsn,
+    commit_lsn: Lsn,
    /// combined hot standby feedback from all replicas
    hs_feedback: HotStandbyFeedback,
 }
@@ -72,7 +72,7 @@ impl SharedState {
        };

        Ok(Self {
-            notified_commit_lsn: Lsn(0),
+            commit_lsn: Lsn(0),
            sk: SafeKeeper::new(Lsn(flush_lsn), tli, storage, state),
            hs_feedback: HotStandbyFeedback {
                ts: 0,
@@ -112,7 +112,7 @@ impl SharedState {
        }
        match opts.open(&control_file_path) {
            Ok(mut file) => {
-                // Lock file to prevent two or more active safekeepers
+                // Lock file to prevent two or more active wal_acceptors
                match file.try_lock_exclusive() {
                    Ok(()) => {}
                    Err(e) => {
@@ -186,7 +186,7 @@ impl Timeline {
    pub fn wait_for_lsn(&self, lsn: Lsn) -> Lsn {
        let mut shared_state = self.mutex.lock().unwrap();
        loop {
-            let commit_lsn = shared_state.notified_commit_lsn;
+            let commit_lsn = shared_state.commit_lsn;
            // This must be `>`, not `>=`.
            if commit_lsn > lsn {
                return commit_lsn;
@@ -198,8 +198,8 @@ impl Timeline {
    // Notify caught-up WAL senders about new WAL data received
    pub fn notify_wal_senders(&self, commit_lsn: Lsn) {
        let mut shared_state = self.mutex.lock().unwrap();
-        if shared_state.notified_commit_lsn < commit_lsn {
-            shared_state.notified_commit_lsn = commit_lsn;
+        if shared_state.commit_lsn < commit_lsn {
+            shared_state.commit_lsn = commit_lsn;
            self.cond.notify_all();
        }
    }
@@ -289,7 +289,7 @@ lazy_static! {
 }

 /// A zero-sized struct used to manage access to the global timelines map.
-pub struct GlobalTimelines;
+struct GlobalTimelines;

 impl GlobalTimelines {
    /// Get a timeline with control file loaded from the global TIMELINES map.
@@ -337,14 +337,14 @@ impl Storage for FileStorage {
        Ok(())
    }

-    fn write_wal(&mut self, server: &ServerInfo, startpos: Lsn, buf: &[u8]) -> Result<()> {
+    fn write_wal(&mut self, s: &SafeKeeperState, startpos: Lsn, buf: &[u8]) -> Result<()> {
        let mut bytes_left: usize = buf.len();
        let mut bytes_written: usize = 0;
        let mut partial;
        let mut start_pos = startpos;
        const ZERO_BLOCK: &[u8] = &[0u8; XLOG_BLCKSZ];
-        let wal_seg_size = server.wal_seg_size as usize;
-        let ztli = server.ztli;
+        let wal_seg_size = s.server.wal_seg_size as usize;
+        let ztli = s.server.ztli;

        /* Extract WAL location for this block */
        let mut xlogoff = start_pos.segment_offset(wal_seg_size) as usize;
@@ -365,7 +365,7 @@ impl Storage for FileStorage {
            /* Open file */
            let segno = start_pos.segment_number(wal_seg_size);
            // note: we basically don't support changing pg timeline
-            let wal_file_name = XLogFileName(server.tli, segno, wal_seg_size);
+            let wal_file_name = XLogFileName(s.server.tli, segno, wal_seg_size);
            let wal_file_path = self
                .conf
                .data_dir
--- a/walkeeper/src/wal_service.rs
+++ b/walkeeper/src/wal_service.rs
@@ -12,7 +12,13 @@ use crate::WalAcceptorConf;
 use zenith_utils::postgres_backend::{AuthType, PostgresBackend};

 /// Accept incoming TCP connections and spawn them into a background thread.
-pub fn thread_main(conf: WalAcceptorConf, listener: TcpListener) -> Result<()> {
+pub fn thread_main(conf: WalAcceptorConf) -> Result<()> {
+    info!("Starting wal acceptor on {}", conf.listen_addr);
+    let listener = TcpListener::bind(conf.listen_addr.clone()).map_err(|e| {
+        error!("failed to bind to address {}: {}", conf.listen_addr, e);
+        e
+    })?;
+
    loop {
        match listener.accept() {
            Ok((socket, peer_addr)) => {
@@ -35,8 +41,8 @@ fn handle_socket(socket: TcpStream, conf: WalAcceptorConf) -> Result<()> {
    socket.set_nodelay(true)?;

    let mut conn_handler = SendWalHandler::new(conf);
-    let pgbackend = PostgresBackend::new(socket, AuthType::Trust, None, false)?;
-    // libpq replication protocol between safekeeper and replicas/pagers
+    let pgbackend = PostgresBackend::new(socket, AuthType::Trust, None)?;
+    // libpq replication protocol between wal_acceptor and replicas/pagers
    pgbackend.run(&mut conn_handler)?;

    Ok(())
--- a/zenith/src/main.rs
+++ b/zenith/src/main.rs
@@ -88,12 +88,7 @@ fn main() -> Result<()> {
        )
        .subcommand(SubCommand::with_name("status"))
        .subcommand(SubCommand::with_name("start").about("Start local pageserver"))
-        .subcommand(SubCommand::with_name("stop").about("Stop local pageserver")
-                    .arg(Arg::with_name("immediate")
-                    .help("Don't flush repository data at shutdown")
-                    .required(false)
-                    )
-        )
+        .subcommand(SubCommand::with_name("stop").about("Stop local pageserver"))
        .subcommand(SubCommand::with_name("restart").about("Restart local pageserver"))
        .subcommand(
            SubCommand::with_name("pg")
@@ -201,12 +196,10 @@ fn main() -> Result<()> {
            }
        }

-        ("stop", Some(stop_match)) => {
+        ("stop", Some(_sub_m)) => {
            let pageserver = PageServerNode::from_env(&env);

-            let immediate = stop_match.is_present("immediate");
-
-            if let Err(e) = pageserver.stop(immediate) {
+            if let Err(e) = pageserver.stop() {
                eprintln!("pageserver stop failed: {}", e);
                exit(1);
            }
@@ -215,8 +208,7 @@ fn main() -> Result<()> {
        ("restart", Some(_sub_m)) => {
            let pageserver = PageServerNode::from_env(&env);

-            //TODO what shutdown strategy should we use here?
-            if let Err(e) = pageserver.stop(false) {
+            if let Err(e) = pageserver.stop() {
                eprintln!("pageserver stop failed: {}", e);
                exit(1);
            }
--- a/zenith_metrics/Cargo.toml
+++ b/zenith_metrics/Cargo.toml
@@ -7,4 +7,3 @@ edition = "2018"
 prometheus = {version = "0.12", default_features=false} # removes protobuf dependency
 libc = "0.2"
 lazy_static = "1.4"
-once_cell = "1.8.0"
--- a/zenith_metrics/src/lib.rs
+++ b/zenith_metrics/src/lib.rs
@@ -3,10 +3,7 @@
 //! Otherwise, we might not see all metrics registered via
 //! a default registry.
 use lazy_static::lazy_static;
-use once_cell::race::OnceBox;
 pub use prometheus::{exponential_buckets, linear_buckets};
-pub use prometheus::{register_gauge, Gauge};
-pub use prometheus::{register_gauge_vec, GaugeVec};
 pub use prometheus::{register_histogram, Histogram};
 pub use prometheus::{register_histogram_vec, HistogramVec};
 pub use prometheus::{register_int_counter, IntCounter};
@@ -23,55 +20,17 @@ pub use wrappers::{CountedReader, CountedWriter};
 /// Metrics gathering is a relatively simple and standalone operation, so
 /// it might be fine to do it this way to keep things simple.
 pub fn gather() -> Vec<prometheus::proto::MetricFamily> {
-    update_rusage_metrics();
+    update_io_metrics();
    prometheus::gather()
 }

-static COMMON_METRICS_PREFIX: OnceBox<&str> = OnceBox::new();
-
-/// Sets a prefix which will be used for all common metrics, typically a service
-/// name like 'pageserver'. Should be executed exactly once in the beginning of
-/// any executable which uses common metrics.
-pub fn set_common_metrics_prefix(prefix: &'static str) {
-    // Not unwrap() because metrics may be initialized after multiple threads have been started.
-    COMMON_METRICS_PREFIX
-        .set(prefix.into())
-        .unwrap_or_else(|_| {
-            eprintln!(
-                "set_common_metrics_prefix() was called second time with '{}', exiting",
-                prefix
-            );
-            std::process::exit(1);
-        });
-}
-
-/// Prepends a prefix to a common metric name so they are distinguished between
-/// different services, see <https://github.com/zenithdb/zenith/pull/681>
-/// A call to set_common_metrics_prefix() is necessary prior to calling this.
-pub fn new_common_metric_name(unprefixed_metric_name: &str) -> String {
-    // Not unwrap() because metrics may be initialized after multiple threads have been started.
-    format!(
-        "{}_{}",
-        COMMON_METRICS_PREFIX.get().unwrap_or_else(|| {
-            eprintln!("set_common_metrics_prefix() was not called, but metrics are used, exiting");
-            std::process::exit(1);
-        }),
-        unprefixed_metric_name
-    )
-}
-
 lazy_static! {
    static ref DISK_IO_BYTES: IntGaugeVec = register_int_gauge_vec!(
-        new_common_metric_name("disk_io_bytes"),
+        "pageserver_disk_io_bytes",
        "Bytes written and read from disk, grouped by the operation (read|write)",
        &["io_operation"]
    )
    .expect("Failed to register disk i/o bytes int gauge vec");
-    static ref MAXRSS_KB: IntGauge = register_int_gauge!(
-        new_common_metric_name("maxrss_kb"),
-        "Memory usage (Maximum Resident Set Size)"
-    )
-    .expect("Failed to register maxrss_kb int gauge");
 }

 // Records I/O stats in a "cross-platform" way.
@@ -83,7 +42,7 @@ lazy_static! {
 // We know the size of the block, so we can determine the I/O bytes out of it.
 // The value might be not 100% exact, but should be fine for Prometheus metrics in this case.
 #[allow(clippy::unnecessary_cast)]
-fn update_rusage_metrics() {
+fn update_io_metrics() {
    let rusage_stats = get_rusage_stats();

    const BYTES_IN_BLOCK: i64 = 512;
@@ -93,7 +52,6 @@ fn update_rusage_metrics() {
    DISK_IO_BYTES
        .with_label_values(&["write"])
        .set(rusage_stats.ru_oublock * BYTES_IN_BLOCK);
-    MAXRSS_KB.set(rusage_stats.ru_maxrss);
 }

 fn get_rusage_stats() -> libc::rusage {
--- a/zenith_utils/Cargo.toml
+++ b/zenith_utils/Cargo.toml
@@ -18,9 +18,12 @@ serde = { version = "1.0", features = ["derive"] }
 serde_json = "1"
 thiserror = "1.0"
 tokio = "1.11"
-tracing = "0.1"
-tracing-log = "0.1"
-tracing-subscriber = "0.2"
+
+slog-async = "2.6.0"
+slog-stdlog = "4.1.0"
+slog-scope = "4.4.0"
+slog-term = "2.8.0"
+slog = "2.7.0"

 zenith_metrics = { path = "../zenith_metrics" }
 workspace_hack = { path = "../workspace_hack" }
@@ -35,4 +38,3 @@ rustls-split = "0.2.1"
 hex-literal = "0.3"
 bytes = "1.0"
 webpki = "0.21"
-tempfile = "3.2"
--- a/zenith_utils/src/accum.rs
+++ b/zenith_utils/src/accum.rs
@@ -1,33 +0,0 @@
-/// A helper to "accumulate" a value similar to `Iterator::reduce`, but lets you
-/// feed the accumulated values by calling the 'accum' function, instead of having an
-/// iterator.
-///
-/// For example, to calculate the smallest value among some integers:
-///
-/// ```
-/// use zenith_utils::accum::Accum;
-///
-/// let values = [1, 2, 3];
-///
-/// let mut min_value: Accum<u32> = Accum(None);
-/// for new_value in &values {
-///     min_value.accum(std::cmp::min, *new_value);
-/// }
-///
-/// assert_eq!(min_value.0.unwrap(), 1);
-/// ```
-pub struct Accum<T>(pub Option<T>);
-impl<T: Copy> Accum<T> {
-    pub fn accum<F>(&mut self, func: F, new_value: T)
-    where
-        F: FnOnce(T, T) -> T,
-    {
-        // If there is no previous value, just store the new value.
-        // Otherwise call the function to decide which one to keep.
-        self.0 = Some(if let Some(accum) = self.0 {
-            func(accum, new_value)
-        } else {
-            new_value
-        });
-    }
-}
--- a/zenith_utils/src/bin_ser.rs
+++ b/zenith_utils/src/bin_ser.rs
@@ -94,12 +94,9 @@ pub fn le_coder() -> impl Options {

 /// Binary serialize/deserialize helper functions (Big Endian)
 ///
-pub trait BeSer {
+pub trait BeSer: Serialize + DeserializeOwned {
    /// Serialize into a byte slice
-    fn ser_into_slice(&self, mut b: &mut [u8]) -> Result<(), SerializeError>
-    where
-        Self: Serialize,
-    {
+    fn ser_into_slice(&self, mut b: &mut [u8]) -> Result<(), SerializeError> {
        // &mut [u8] implements Write, but `ser_into` needs a mutable
        // reference to that. So we need the slightly awkward "mutable
        // reference to a mutable reference.
@@ -110,28 +107,19 @@ pub trait BeSer {
    ///
    /// This is useful for most `Write` types except `&mut [u8]`, which
    /// can more easily use [`ser_into_slice`](Self::ser_into_slice).
-    fn ser_into<W: Write>(&self, w: &mut W) -> Result<(), SerializeError>
-    where
-        Self: Serialize,
-    {
+    fn ser_into<W: Write>(&self, w: &mut W) -> Result<(), SerializeError> {
        be_coder().serialize_into(w, &self).map_err(|e| e.into())
    }

    /// Serialize into a new heap-allocated buffer
-    fn ser(&self) -> Result<Vec<u8>, SerializeError>
-    where
-        Self: Serialize,
-    {
+    fn ser(&self) -> Result<Vec<u8>, SerializeError> {
        be_coder().serialize(&self).map_err(|e| e.into())
    }

    /// Deserialize from the full contents of a byte slice
    ///
    /// See also: [`BeSer::des_prefix`]
-    fn des(buf: &[u8]) -> Result<Self, DeserializeError>
-    where
-        Self: DeserializeOwned,
-    {
+    fn des(buf: &[u8]) -> Result<Self, DeserializeError> {
        be_coder()
            .deserialize(buf)
            .or(Err(DeserializeError::BadInput))
@@ -143,10 +131,7 @@ pub trait BeSer {
    /// type, but does not guarantee that the entire slice is used.
    ///
    /// See also: [`BeSer::des`]
-    fn des_prefix(buf: &[u8]) -> Result<Self, DeserializeError>
-    where
-        Self: DeserializeOwned,
-    {
+    fn des_prefix(buf: &[u8]) -> Result<Self, DeserializeError> {
        be_coder()
            .allow_trailing_bytes()
            .deserialize(buf)
@@ -154,10 +139,7 @@ pub trait BeSer {
    }

    /// Deserialize from a reader
-    fn des_from<R: Read>(r: &mut R) -> Result<Self, DeserializeError>
-    where
-        Self: DeserializeOwned,
-    {
+    fn des_from<R: Read>(r: &mut R) -> Result<Self, DeserializeError> {
        be_coder().deserialize_from(r).map_err(|e| e.into())
    }

@@ -165,22 +147,16 @@ pub trait BeSer {
    ///
    /// Note: it may be faster to serialize to a buffer and then measure the
    /// buffer length, than to call `serialized_size` and then `ser_into`.
-    fn serialized_size(&self) -> Result<u64, SerializeError>
-    where
-        Self: Serialize,
-    {
+    fn serialized_size(&self) -> Result<u64, SerializeError> {
        be_coder().serialized_size(self).map_err(|e| e.into())
    }
 }

 /// Binary serialize/deserialize helper functions (Little Endian)
 ///
-pub trait LeSer {
+pub trait LeSer: Serialize + DeserializeOwned {
    /// Serialize into a byte slice
-    fn ser_into_slice(&self, mut b: &mut [u8]) -> Result<(), SerializeError>
-    where
-        Self: Serialize,
-    {
+    fn ser_into_slice(&self, mut b: &mut [u8]) -> Result<(), SerializeError> {
        // &mut [u8] implements Write, but `ser_into` needs a mutable
        // reference to that. So we need the slightly awkward "mutable
        // reference to a mutable reference.
@@ -191,28 +167,19 @@ pub trait LeSer {
    ///
    /// This is useful for most `Write` types except `&mut [u8]`, which
    /// can more easily use [`ser_into_slice`](Self::ser_into_slice).
-    fn ser_into<W: Write>(&self, w: &mut W) -> Result<(), SerializeError>
-    where
-        Self: Serialize,
-    {
+    fn ser_into<W: Write>(&self, w: &mut W) -> Result<(), SerializeError> {
        le_coder().serialize_into(w, &self).map_err(|e| e.into())
    }

    /// Serialize into a new heap-allocated buffer
-    fn ser(&self) -> Result<Vec<u8>, SerializeError>
-    where
-        Self: Serialize,
-    {
+    fn ser(&self) -> Result<Vec<u8>, SerializeError> {
        le_coder().serialize(&self).map_err(|e| e.into())
    }

    /// Deserialize from the full contents of a byte slice
    ///
    /// See also: [`LeSer::des_prefix`]
-    fn des(buf: &[u8]) -> Result<Self, DeserializeError>
-    where
-        Self: DeserializeOwned,
-    {
+    fn des(buf: &[u8]) -> Result<Self, DeserializeError> {
        le_coder()
            .deserialize(buf)
            .or(Err(DeserializeError::BadInput))
@@ -224,10 +191,7 @@ pub trait LeSer {
    /// type, but does not guarantee that the entire slice is used.
    ///
    /// See also: [`LeSer::des`]
-    fn des_prefix(buf: &[u8]) -> Result<Self, DeserializeError>
-    where
-        Self: DeserializeOwned,
-    {
+    fn des_prefix(buf: &[u8]) -> Result<Self, DeserializeError> {
        le_coder()
            .allow_trailing_bytes()
            .deserialize(buf)
@@ -235,10 +199,7 @@ pub trait LeSer {
    }

    /// Deserialize from a reader
-    fn des_from<R: Read>(r: &mut R) -> Result<Self, DeserializeError>
-    where
-        Self: DeserializeOwned,
-    {
+    fn des_from<R: Read>(r: &mut R) -> Result<Self, DeserializeError> {
        le_coder().deserialize_from(r).map_err(|e| e.into())
    }

@@ -246,18 +207,14 @@ pub trait LeSer {
    ///
    /// Note: it may be faster to serialize to a buffer and then measure the
    /// buffer length, than to call `serialized_size` and then `ser_into`.
-    fn serialized_size(&self) -> Result<u64, SerializeError>
-    where
-        Self: Serialize,
-    {
+    fn serialized_size(&self) -> Result<u64, SerializeError> {
        le_coder().serialized_size(self).map_err(|e| e.into())
    }
 }

-// Because usage of `BeSer` or `LeSer` can be done with *either* a Serialize or
-// DeserializeOwned implementation, the blanket implementation has to be for every type.
-impl<T> BeSer for T {}
-impl<T> LeSer for T {}
+impl<T> BeSer for T where T: Serialize + DeserializeOwned {}
+
+impl<T> LeSer for T where T: Serialize + DeserializeOwned {}

 #[cfg(test)]
 mod tests {
--- a/zenith_utils/src/connstring.rs
+++ b/zenith_utils/src/connstring.rs
@@ -29,24 +29,24 @@ mod tests {

    #[test]
    fn test_connection_host_port() {
-        let config: Config = "postgresql://no_user@127.0.0.1:64000/no_db"
+        let config: Config = "postgresql://no_user@localhost:64000/no_db"
            .parse()
            .unwrap();
        assert_eq!(
            connection_host_port(&config),
-            ("127.0.0.1".to_owned(), 64000)
+            ("localhost".to_owned(), 64000)
        );
    }

    #[test]
    #[should_panic(expected = "only one pair of host and port is supported in connection string")]
    fn test_connection_host_port_multiple_ports() {
-        let config: Config = "postgresql://no_user@127.0.0.1:64000,127.0.0.1:64001/no_db"
+        let config: Config = "postgresql://no_user@localhost:64000,localhost:64001/no_db"
            .parse()
            .unwrap();
        assert_eq!(
            connection_host_port(&config),
-            ("127.0.0.1".to_owned(), 64000)
+            ("localhost".to_owned(), 64000)
        );
    }
 }
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Konstantin Knizhnik	458ca82d75	Bump postgres version	2021-09-27 12:08:55 +03:00
Konstantin Knizhnik	0dface838d	Bump postgres version	2021-09-24 19:21:16 +03:00
Konstantin Knizhnik	f7878c5157	Bump postgres version	2021-09-24 19:10:18 +03:00
Konstantin Knizhnik	53c870b8e7	Store file cache outside pgdata dir	2021-09-24 19:07:30 +03:00
Konstantin Knizhnik	c23b65914e	Sete zenith.file_cache_size parameter	2021-09-24 18:19:41 +03:00
Konstantin Knizhnik	b51d3f6b2b	Revert "Save page received from page server in local file cache" This reverts commit `137472db91`.	2021-09-24 18:04:21 +03:00
Konstantin Knizhnik	137472db91	Save page received from page server in local file cache	2021-09-24 15:15:54 +03:00
Konstantin Knizhnik	f817985f2b	Do not throw error on attempt to drop unexisted relation	2021-09-23 12:11:06 +03:00
Konstantin Knizhnik	2e94e1428e	Remove import_timeline_wal function	2021-09-23 12:11:06 +03:00
Konstantin Knizhnik	9a486ca109	Do not write WAL at pageserver	2021-09-23 12:11:06 +03:00